Maximizing prescriptive analytics to reach sustainability
Organizations seeking to retain prescriptive information and use it for future events can store this data directly in a self-service advanced analytics solution.
Sustainability is a goal for a variety of noble reasons for companies in many different industries. Particularly in oil and gas processing, sulfur recovery needs to be properly managed to respect the environment as well as to improve production.
Sulfur is harmful to the environment and the EPA regulates this; sometimes imposing fines in the millions of dollars. These possible fines go beyond any potential losses to production and/or quality. Sulfur recovery units are an essential means for engineers to track process behavior. When an anomaly occurs in sulfur recovery, prescriptive analytics solutions lean on root cause analysis to guide the solution, gather intel, and help engineers to pinpoint conditions when an event may occur.
But what if the engineer who has the experience is not the one monitoring the process, or happens to leave the organization? Such a gap in knowledge could be detrimental to the final product and ultimately the organization’s bottom line. It also could have a serious effect on a company’s sustainability objectives as sulfur makes its way into finished products.
Organizations seeking to retain prescriptive information and use it for future events can store this data directly in a self-service advanced analytics solution. They can include these specific recommendations for process experts to use in conjunction with time-series data. Engineers throughout the company can share this information and know what to do when a problem occurs, even before it happens.
This information is a type of contextual data. Contextual data complements time-series data and helps to improve analysis. Maintenance logs, laboratory information management systems, and information residing in third-party solutions are examples of contextual data. They often are locked away in separate applications throughout an organization, which creates data silos. Eliminating these silos and making the information available to anyone empowers all process experts to make sustainable decisions.
Here’s how placing data in context using a self-service advanced analytics solution can help an organization achieve sustainability goals using prescriptive analytics.
Overcoming an anomaly
The most common way of producing sulfur is the multi-stage Claus process. A simplified flowchart of this process is shown in figure 1.
The process has multiple stages because of its chemical equilibrium. A feed gas containing a high concentration of H2S must be pre-heated using steam. The gas then enters the first converter, where it undergoes partial conversion toward pure sulfur. Part of the sulfur is removed from the gas stream before entering the second converter. In the final stage, the pure sulfur is removed and the remaining gas containing small amounts of H2S goes downstream to a treatment plant.
Engineers at a European refinery were viewing the recent trends of a sulfur recovery unit on a dashboard in the refinery’s advanced analytics solution. They noticed that the amount of sulfur produced in the recovery unit, represented as tons per day, had dropped significantly. Process experts soon learned the sulfur yield also was declining, which indicated a lower conversion in the plant. As a result, the higher amount of H2S remaining in the gas stream put too much load on the gas treatment plant.
Several things could cause this problem. Engineers might find an issue with the catalyst in the converters, temperature of the converters, temperatures of the condensers, or the control systems themselves. Process experts used the advanced analytics solution to determine the root cause of the low conversion. They learned that a malfunction in a temperature-control valve upstream led to a low inlet temperature in the first converter. There is a clear correlation between sulfur recovery (x-axis) and the converter inlet temperature (y-axis), where the more recent points in orange indicate a lower recovery correlated with a lower converter inlet temperature.
Process experts wanted to save the findings and make them available to colleagues who encounter the same problem. They added the information as contextual data in their advanced analytics solution.
Analyzing time-series and contextual data often has required the help of a data scientist. The classic data science approach includes developing data models so a data scientist can crunch numbers. But data scientists are scarce, and this approach is expensive and time-consuming. It also has other limitations.
Engineers and data scientists do not speak the same language, which causes a communication gap between them. Furthermore, the data scientist may not be familiar with the process that needs to be analyzed. Process experts could spend a significant amount of time explaining the process to the data scientist so he or she can perform the analysis.
A self-service advanced analytics solution enables engineers to run the analysis themselves without the help of a data scientist. Self-service solutions do not require data modeling. They foster the analysis of time-series data along with contextual data. Engineers do not have to keep separate log files in separate systems, which prevents the buildup of data silos.
An advanced analytics solution reduces the need for specialized skills. It also eliminates the need to understand computer programming languages and know other statistical tools. Finally, engineers will find that determining a single root cause is far easier in an advanced analytics solution when an anomaly could correlate with multiple tags.
Ideally, the insights learned during the root cause analysis are captured in the form of prescriptions that can be loaded immediately into the advanced analytics solution. In practice, however, creating the prescription requires some additional work. The following three steps explain the process of storing prescriptions as contextual data.
Step 1: Contextualizing time-series data
In addition to challenges of entering and storing all process data, time-series data has its own limitations. First, it tends to contain a lot of noise that makes the data hard to read. The noise must be filtered out before the data can be used for analysis. Second, it contains periods of normal operation as well as abnormal operation. Normal operating timeframes can be used to identify a golden fingerprint, or the parameters necessary for ideal process behavior. When evaluating for anomalies, however, normal periods of operation also need to be removed.
Therefore, the first question should be, “Is the process running normally?” For a simple answer, engineers can use a Gantt chart to highlight periods of low production in the time-series data. In the case of Figure 3, only the areas in orange are interesting for the analysis.
For a quick analysis, this graph provides a simple solution to monitor periods when the process was not operational. Engineers can perform this exercise for all the relevant time-series data.
Step 2: Assembling relevant contextual views
Once the time-series data has been structured and the contextual data has been loaded, process experts can use the advanced analytics solution to apply filters. These make the contextual views.
In the next phase, engineers can see several tags that have been created for the various options regarding sulfur recovery: Normal, High, and Low. The corresponding Gantt chart lets engineers quickly see how long each of these events has lasted. A longer bar means the specific event lasted longer.
Data comes from the monitor previously set up during Step 1. When production reaches a certain threshold, a trigger in the monitor creates a context item. This item, in turn, appears in the Gantt chart. Process experts can save and share these reports throughout the organization.
Step 3: Applying prescriptive analytics
The report created during Step 2 needs to be extended before applying prescriptive analytics. Extending the report includes adding tags that note measurements of sulfur yield, H2S concentration, chemical converter parameters, and so forth. Figure 3 shows a Gantt chart that contains all relevant contextual data necessary to perform the root analysis in earlier steps. For example, the left column shows extra parameters, such as sulfur converter 1, outlet measurement of H2S gas, sulfur re-heater, and others.
In the advanced analytics solution, the reports associated with these events provide descriptive information that explain what happened. The time-series data also provides the answer to the question, “when.” Furthermore, it provides a list of things to check when the event occurred to determine why.
For example, in the case of low sulfur production, process experts see that the first thing to check is sulfur yield. If the yield is low, engineers need to intervene only when the alert associated with the monitor has tripped multiple times. If it has, process experts can get a list of instructions and a checklist that will tell them what to look for to correct the anomaly.
Sustainable prescriptive analytics
It makes sense for process experts to use contextual data to prescribe the solution for any anomalies that may occur. Problems can be resolved more effectively by using advanced analytics.
Alongside these production improvements, sustainability can be met to address environmental concerns, including carbon emissions, energy usage, and water/wastewater management.
Putting information in context enables engineers to make data-driven decisions – and in the case of sulfur recovery, perhaps solve some serious environmental issues.