The ongoing rise of the Industrial Internet of Things (IIOT) is spurring the generation of ever-increasing amounts of data at plants. However, as many chemical makers can attest, more data doesn’t necessarily lead to more insights about operations. Uncovering underlying relationships among data to identify trends that affect a plant hasn’t been easy and often has demanded the skills of data scientists. Fortunately, developments in predictive analytics software are enabling engineers on site to make more effective use of data to improve operations. A number of vendors and even an operating company now offer innovative technologies that are producing impressive results.
For instance, TrendMiner, Houston, and Hasselt, Belgium, points to its work with Total Refining and Chemical. This began with a pilot project at Total’s plant in Antwerp, Belgium, to use TrendMiner’s predictive analytics to optimize process performance.
“The pilot OSIsoft real-time process data infrastructure with the self-service process and asset analytics software of TrendMiner showed significant benefits in all four industrial priorities for Total: reduce safety risks, improve asset availability, reduce operational costs and increase energy efficiency. The 45 experts involved gave it the highest average rate for any software tool ever assessed by Total on user friendliness and capabilities to improve overall plant performance,” says TrendMiner CEO Bert Baeck.
The TrendMiner support team involved in the project reported a three-to-five times return on investment in the first year. Based on this success, Total has started to roll out the software to its entire refining and chemicals business unit.
The analysis software relies on a patent-pending pattern recognition technology designed to find in seconds similar (multivariate) process behavior in up to a decade’s worth of data. This speed of analyzing process behavior is just the starting point for identifying root causes; a range of tags can be instantly assessed to find influencing factors for anomalies.
TrendMiner is unique, according to Baeck, because of its second patent-pending technology to predict process behavior based on behavior in the past. Rather than creating predictive analytics models, the process engineer can create fingerprints to monitor the real-time process performance. These fingerprints are used to send early warnings, based on user-defined rules, in case of deviations but also allow comparing actual and ideal behavior. In the case of a deviation, the notifications enable field operators or control room personnel to take appropriate, recommended action. Fingerprints also can serve to capture interesting events for future analysis to further optimize plant performance.
The self-service approach has many benefits over traditional data modeling, too, especially in terms of time demands, contends Baeck: “Subject matter experts can easily verify process anomalies based on their expertise via graphical representation of the process behavior for a specific tag, so avoiding the need for complex data modeling.”
He cites another process example at a plant that was having problems with the cooling phase in a reactor during polymer production due to heat exchanger fouling. Monitoring the fouling was extremely difficult because the reactor produces a variety of grades of polymer, each requiring a different recipe. The plant faced the classic dilemma: scheduling maintenance too early leads to unwarranted reactor downtime while leaving it too late degrades performance and poses potential risks such as insufficient cooling capacity in emergency situations.
So, the plant deployed a monitor to check on the cooling time for the most frequent products. The start of an increase in the duration of the cooling phase prompts a warning to its engineers, who then can schedule timely maintenance. This has resulted in extended asset availability, reduced operational and maintenance costs, and decreased safety risk.
No Data Movement
Total’s experience with predictive analytics comes as no surprise to Michael Risse, vice president of Seeq Corp., Seattle. Chemical and pharmaceutical companies seem to appreciate and value improvements they can get more than other firms and so have a greater willingness to adopt the technology, he says.
“This is because they already have a lot of context in their batch historians to complement their process data,” notes Seeq analytics engineer Brian Crandall.
Seeq’s software application workbench sits in the user’s browser and is designed to be very easy to work with. The Seeq server only stores Seeq data, for example workbooks and worksheets, but no customer data. “This is part of our low touch model: we only connect to the customer data. It’s a differentiator because many approaches to predictive analytics require data movement, to a data lake or the cloud for example, but we do not. Our philosophy is “your data is fine; leave it where it is,” Crandall adds.
So the keys to Seeq’s technology are the connectors linking the Seeq server to process data inputs and historian data. These are designed to add depth to the data already existing in historians (Figure 1).
Users can employ a whole range of tools such as value search, pattern search, composite condition, signal from condition and more to generate capsules (user-defined slices of time with a unique identifier).
“These are unique to Seeq and designed to make it very simple to import any data you might want. It is being able to access any data that is the critical step here and gives the software its predictive power. The important thing to note here is that capsules are super easy to use and can be defined in so many ways, for example from searches or from a batch, etc.,” stresses Risse.
This ease of use and depth of data analysis was critical in a recent project with Abbott in the U.S. that focused on asset optimization and making additional salable product without increasing variable costs.
Seeq used its advanced analytics to leverage the existing plant optimization system and historian. This revealed two key opportunities for improvement: in the clean-in-place (CIP) process, and in product giveaway in packaging.
Each CIP routine consists of multiple circuits through a five-step process involving caustic circulation, caustic rinse, acid circulation, acid rinse and then final rinse.
Seeq created a process model from the data and applied it to the different circuits. This indicated that cleaning went on too long. Based on this, Abbot reduced CIP cycle time by an average of 20 minutes — with an aggregate impact of three extra hours of production each month without the need for extra maintenance.
In assessing product filling on the packaging line, the software again leveraged the historian to acquire high-speed packaging data, analyses and reporting.
Seeq’s analytics pinpointed how to identify and control variability in the filler equipment. This allowed Abbott to set the filling-weight set point closer to its target, totalize the giveaway amount across shifts to increase staff awareness, and also carry out continuous monitoring of filler-head actual weight distribution to aid with condition-based maintenance.
The latest release of Seeq’s software, R17, has new tools designed to add even more depth to its predictive analysis capabilities. For example, the prediction tool speeds up model creation. The reference profile tool characterizes a signal’s intended behavior — or its anomalous behavior — for purposes such as statistical process control and boundary definitions. The new journal tool is proving very popular with engineers, says Risse, because it allows them to write and take notes as they work. These captured insights can be saved, searched and shared — broadening the impact of data analyses within plants, sites and companies.
Soda ash producer Ciner Resources, Green River, Wyo., turned to pattern recognition software from Falkonry, Santa Clara, Calif., to help overcome unplanned outage issues with its Vertimill grinding plant.
“Vertimill downtime affects 60% of our hourly production,” notes Ciner CIO Scott Schemmel. “Falkonry pattern recognition technology helped us solve a complex, costly industrial business problem fast and has shown potential to significantly reduce unplanned downtime. The Falkonry technology was impressive for how quickly we could set up and leverage it,” he adds.
Falkonry combines signal processing and machine learning to discover, recognize and predict the condition of operating assets and production processes.
“Explaining our manufacturing process to someone outside the refinery environment is difficult. It was important to us that Falkonry enabled our own subject matter experts to solve our problems — we didn’t have to teach anyone outside the plant our process. We just gathered our own data and used Falkonry ourselves. Best of all, the relevant patterns in our data became recognizable by Falkonry within a matter of days,” notes Ciner smart plant lead Jolene Baker.
Ease of use also is one of the major drivers for EDM for Manufacturing, new predictive analytics software from Sightline Systems, Fairfax, Va. It aims to help chemical and other manufacturers take advantage of IIOT system and device data streams, enabling them proactively to address costly challenges that can develop throughout the production process.
The software uses advanced machine-learning techniques to significantly simplify predicting future operational performance, says the company. By reviewing collected historical data, EDM for Manufacturing learns the best statistical model to provide a reliable forecast of future behavior.
“The advanced analytics capability within EDM for Manufacturing rapidly correlates thousands of data points at a sub-second level and quickly presents the information in multiple, easy-to-understand formats, enabling industrial operators to automate processes and zero in on a potential issue before it becomes a problem,” explains president Brandon Witte.
Some suppliers are collaborating on predictive analytics offerings. For example, Dover Energy Automation (DEA), The Woodlands, Texas, is teaming up with cognitive computing analytics specialist SparkCognition, Austin, Texas. Essentially, SparkCognition is providing a machine-learning engine for DEA to integrate into the intelligence platform of its hardware.
The combination of SparkCognition’s capabilities in artificial intelligence technology, cyber security and software development and DEA’s deep domain knowledge in condition monitoring and asset optimization will create what DEA president Ali Raza describes as a robust IIOT ecosystem that should enable customers to deal with previously unsolvable problems.
DEA also is collaborating with Honeywell Process Solutions, Houston, as part of the latter’s Inspire program. It brings together a community of technology users and providers — including customers, equipment vendors, process licensors and Honeywell experts — that will jointly develop technologies for myriad operational challenges.
According to Seeq’s Risse, the widespread recognition in the chemical industry of the need for a new tool to carry out ad hoc investigations of processes — and among automation vendors that they have a gap for a better application experience — is driving partnerships with and between companies such as Seeq, Honeywell, Schneider Electric, OSIsoft, Siemens, Emerson, GE and Aspen Technology.
And the efforts aren’t restricted to vendors. For instance, Air Products, Allentown, Pa., developed its own predictive-monitoring and fault-diagnostic platform called ProcessMD. Today, the patented web-based software monitors hundreds of the company’s plants and thousands of fixed and moving pieces of equipment within them. Air Products estimates the platform contributes multi-million-dollar savings at the sites in annual cost avoidance and productivity benefits. Now, the company is planning to offer customers ProcessMD as a Platform as a Service (PaaS) for real-time asset performance management and diagnostics. (For more on the platform, see “Predictive Analytics Capture Heartbeat of the Plant”).
Seán Ottewell is Chemical Processing's Editor at Large. You can email him at firstname.lastname@example.org.