By Seeing Patterns, Analytics Software Models Improve Process Control

Translating raw data into usable information has been a difficult task for industrial companies. Using TrendMiner, a self-service industrial data analytics software, Sitech was able to overcome this problem.

Advanced polymers, synthetic rubber, fertilizers, metallocene plastomers, polyethelenes, and engineering thermoplastics are among the products made at the Chemelot campus.
Advanced polymers, synthetic rubber, fertilizers, metallocene plastomers, polyethelenes, and engineering thermoplastics are among the products made at the Chemelot campus.

Advanced pattern analysis may sound futuristic, but it’s at work today, helping industries identify and correct process variations that disrupt operations and increase costs. The objective: zero surprises and no unplanned process downtime. Deploying TrendMiner, a self-service industrial data analytics software, has allowed Sitech to achieve those objectives for companies like DSM, Borealis, Arlanxeo, Sabic, OCI Nitrogen, and many others with plants at the Chemelot chemical industry campus in the southern Netherlands.

As a technology partner for Chemelot-based companies, Sitech provides services such as maintenance, technology improvements, and advanced process control (APC) to help customers optimize processes and asset performance. It uses TrendMiner’s analytical tools to predict the performance and conditions of assets at all times, so companies can take early, corrective action and model historical events to replicate best practices in their processes.

While the manufacturing industry owns the most data points in the world—surpassing even Amazon and Google—it is also the furthest behind the curve when it comes to translating data into actual information, according to Marc Pijpers, principal process control engineer at Sitech. Lack of proper tools, insufficient analytics knowledge, minimal embedding of analytics in work processes, unclear economic benefits and insufficient management support, delaying the adoption of digitalization, and better process control are a few of the challenges the manufacturing industry is facing when it comes to data analytics.

“As a result, engineers spend the majority of their time stretching the limits of spreadsheets to acquire and prepare data and visualize problems, instead of actually analyzing the problem and gaining crucial insights from it,” says Pijpers. “There’s also a language barrier between subject matter experts and data scientists when it comes to problem solving. Training process engineers to become data scientists, or data scientists to become process engineers, would be too time-consuming and costly.”

That’s when Pijpers, who has more than 15 years’ experience in designing and implementing basic and advanced process control systems, asked the question: how can we overcome these challenges by exploiting the possibilities of self-service analytics? The aim was to put critical analytical skills and capabilities into the hands of process and control engineers, enabling them to solve problems on their own and passing the more complicated issues on to data scientists.

TrendMiner’s self-service industrial analytics software, which Pijpers calls “a trending client on steroids,” enables Sitech to perform descriptive, diagnostic, and predictive analytics. Which means it helps process engineers gain insight into what happened (descriptive), why and how it happened (diagnostic), and how the situation can be prevented or repeated in the future (predictive).

Sitech engineers use TrendMiner’s self-service analytics software to track patterns in process data.Sitech engineers use TrendMiner’s self-service analytics software to track patterns in process data.

TrendMiner’s advanced pattern search capabilities uses data from historians—such as OSisoft PI and Aspentech IP.21—to compare similar situations, analyze root causes, clean data for APC and Six Sigma to create subsets of relevant data, and monitor important processes to provide early alerts about unwanted variations.

“Big data analytics is a journey,” says Pijpers. “It wasn’t just about buying new software. Process engineers need to learn and adapt to how to use the software and rethink how they handle data. That’s why we decided to launch a pilot project a few years ago using TrendMiner at the Chemelot campus.”

The TrendMiner software helped process engineers solve previously unsolvable issues, verify hypotheses and prove them to be either true or false so they could be addressed or ruled out for the future, use data insights to find new ways to improve performance, and contextualize asset performance for predictive analytics.

One company making different grades of polymers was able to gain more than a 1% increase in revenue over its entire production line by contextualizing asset performance with process data.

The fouling of heat exchangers is also an example of how asset performance is directly related to process behavior. In a reactor with repeated heating and cooling phases, the controlled cooling phase is the most time-consuming process. It’s almost impossible to monitor fouling when the reactor is used for different product grades and a different recipe is required for each grade.

Although fouled heat exchangers increase cooling times, scheduling maintenance too early leads to unwarranted downtime and scheduling too late leads to degraded performance, increased energy consumption, and potential risks.

Sitech set up a monitor to look at the cooling times of the products produced in the highest volumes. By detecting patterns during each production cycle, TrendMiner was able to send engineers a warning to schedule timely maintenance when the duration of the cooling phase increases. As a result, the company gained extended asset availability, was able to better predict maintenance needs, lower costs, and reduce safety risks.

In another case, Sitech process engineers saw the reactor was experiencing thermal stress due to repetitive fast cooling and heating. With TrendMiner, the team set up a way to monitor the number of times the problem occurred to identify how many thermal cycles result in a failure. The software notified the team when the reactor had been thermally stressed. In a matter of minutes, they were able to look back at five years of process history and export valuable search results to perform a lifespan residue analysis.

Sitech was also able to help OCI Nitrogen achieve a more than 5% increase in revenue—representing $2.4 million annually—using TrendMiner’s diagnostic analytics. The Haber-Bosch nitrogen-fixation reaction has been used for over 100 years to make ammonia and other nitrogen-containing compounds from nitrogen and hydrogen. Although it was considered to be a fully optimized process, carbon dioxide peaks were identified in the washing column.

Previous data analysis projects had failed to find the root cause, but TrendMiner’s software helped Sitech search and compare peak periods with normal operation periods. Layer comparisons were used to identify which processes influenced the variables, allowing Sitech to stabilize the operation for increased production.

Diagnostic analytics also provided the process insights that allowed the Sitech team to identify certain peaks in product flows, which were causing unwanted saturation of process sensors and required manual correction for key-performance indicator reporting. By performing influence factor analysis on peaks, they were able to determine the root cause and how highly correlated these events were, as well as the significance of each event. This insight reduced manual labor costs and improved process stability.

“Adapting to new software like TrendMiner often asks for an organizational shift: a new way of looking at technology and data. It enforces alternative thinking and a new way of looking at operational performance using data,” explains Pijpers.

“To prevent process engineers from falling back on traditional ways of working—the well-known Double S curve of innovation—TrendMiner was able to provide us with process engineers on demand who are experienced users,” he adds. “This helped our process engineers analyze issues that are too complex for conventional tools, so they were able to solve problems faster with a shorter learning curve.”

Companies in this article