Do You Trust Your Data?

Control engineers are using myriad tools to ensure that plant data is accurate. Plant data is inherently inconsistent, whether it’s from instrument error or process fluctuation.

Aw 3292 0901 Trust
Yet companies use plant data for operations support, maintenance, planning and optimization. They also send that data up to the enterprise resource planning (ERP) system. So accuracy is critical. Faulty data can result in a hornets’ nest of problems, including poor or unsafe operational decisions.

At the Shell Canada Ltd. refinery in Calgary, Alberta, Canada, plant managers use refinery-wide data reconciliation applications to get a consistent view into production and to identify flow meters that are reading incorrectly. The application uses a model designed by the SimSci-Esscor unit of Invensys Process Systems, in Plano, Texas.

The application runs automatically and produces daily results for Shell’s morning operating meetings. The high-frequency runs were designed to ensure that meter problems are detected close to the event that caused them.

“These online data reconciliation systems are now deployed at several Shell sites worldwide, and provide daily reconciled data,” says Harpreet Gulati, director, hydrocarbon products, at Invensys Process Systems. “The sharp accuracy of the data at Shell Canada gives management and plant personnel the confidence to push the process envelope and make more profitable and safer business decisions.”

Two truths

There are two truths to plant data. One, everyone wants a piece of plant information now—from plant operators to maintenance, asset management and the business side. Two, inaccuracy plagues plant data. Since it’s assumed that data carries inaccuracies, plant operators are turning to data reconciliation tools to improve the data as it travels to various interested parties. Data inaccuracies are getting detected at the sensor level, as the data travels, even within the ERP system itself.

“The accuracy of decisions in an operating environment is directly dependent on the accuracy of the underlying data that is available,” says Gulati. “The accuracy of the data is essential for effective management and accounting, but it also plays into personnel productivity, safety and reliability.” Gulati notes that inaccuracies are a natural part of plant life. “People have learned how to live with the inaccuracies.”

A wide range of data is now passing back and forth from plant manufacturing execution systems (MES) to ERP systems. This includes bills of materials, production data, engineering change management, inventory consumption and shipping information. Given the view that any data being passed likely contains inaccuracies, plants are turning to data reconciliation tools and modeling software to detect corrupt data. “In a distributed environment with 25 different applications, the opportunity for error goes up,” says Jim Kline, business and product manager for collaborative product line at vendor ABB Inc., in Norwalk, Conn. “If all the applications work on the same data entered at one, you improve the data coming to an object or entity in the system.”

Even the best sensors and systems for delivering data carry some degree of errors. “First, people pull measurements and there will be some level of error,” says Invensys’ Gulati. “Secondly, over time, measurements from pressure sensors and temperature sensors degrade. In the best case scenarios, you’ll have an error of plus or minus 2 percent. Typically, it’s 5 percent.”

The more data that’s collected, the greater the possibility for errors. Data inaccuracy can be a particularly difficult problem when data collection comes from devices that are not part of the control system. “With wireless, handheld, barcodes—all to remove paper—we’re getting so much data, we have to ask whether we’re collecting the right data,” says Simon Jacobson, senior research analyst at AMR Research Inc., in Boston. “The biggest challenge is how do we effectively make sure it’s accurate? How do we know if we have the right measurements in place?”

Smooth it out

There are a number of solutions to data inaccuracy. Many sensors have become intelligent enough to police themselves. “The sensors have become more intelligent and are now able to detect more than just out-of-range—they can assign some kind of a quality indicator to the information,” says Keith Jones, program manager for HMI, SCADA and platforms at Wonderware, a Lake Forest, Calif.-based automation software vendor. “In most protocols, there is a point at which the quality of the data is known—you can tell if the data is bad and flag it.”

Control systems and MES can also detect inaccurate data through models and data reconciliation applications that scan data for out-of-norm readings. Sometimes these applications smooth out the data so it presents readings that become in-range averages. These programs can also identify errant sensors that need to be replaced or recalibrated. “We have parameters and ranges for parameters. If the data comes in and it’s outside the range, we can check to see if it’s being sent up correctly,” says Bob Lenich, director of data management at vendor Emerson Process Management, in Austin, Texas. “The control system and MES will take appropriate action because they are smart enough to determine if the data is wrong.”

Control data is often passed through an OPC server (using the OPC open connectivity standard) on its way to delivering information to an ERP system. In most instances, the data is first processed to eliminate inaccuracies. “More and more of the data has already been verified before it gets to OPC,” says Jim Luth, technical director at the OPC Foundation, in Scottsdale, Ariz. “The devices are more and more accurate. And once the quality of the data has been determined, it can pass through as many systems as it needs.”

Filtering data for inaccuracies is not new, but it is moving into new areas. For one, more data is getting shared with more interested parties. Secondly, data sharing is moving down industry tiers to smaller companies. “People who have been doing real-time integration have been filtering the data for accuracy for a long time,” says Joanne Salazar, director of data management services at Emerson. “The big oil and gas plants make sure the data that comes into their models is correct. Now, the smaller and mid-tier chemical and petrochemical plants are starting to do it.”

ERP help?

Now that data is being shared between the plant and the business side, companies are setting up fail-safe programs between the control system and the ERP to make sure data failure in either system will not affect the other system. “If you’re having trouble at the ERP level, you want to make sure the plant can still operate,” says Paul Rauch, senior director at vendor Siemens IT Solutions and Services Inc., in Norwalk, Conn. “And when you’re having difficulties on the plant level, you want to make sure you are able to use the ERP.”

In many instances, the ERP is not configured to scout for data problems and simply accepts data as it arrives. “As you amalgamate data and put it into the ERP, the ERP doesn’t have any idea if it’s good,” says Wonderware’s Jones. “Typically, the ERP believes everything it’s told. If you tell the ERP you’ve made 1,000 pounds of a product and you’ve only consumed 800 pounds of raw materials, the ERP won’t know the difference.”

In other instances, the ERP is actually programmed to take a role in detecting problems with data. “We offer a key functionality to ensure the data collection through exception steps,” says Simo Said, director of global marketing at Germany-based ERP vendor SAP AG. “You can use these models in the ERP system and control the data exchange between the plant system and the ERP system.” The models include a series of eight prediction steps that identify errors in the data.

Some of the ERP data cleansing occurs within the ERP system. “We’ve created a tool so if there’s a data error, it gets passed off to the configurative SAP inbox,” says Siemens’ Rauch. “Here, people can figure out what do to with the errors.”

At other times, the data is reviewed for inaccuracies by ERP tools outside the ERP system itself. “We have a second context outside the ERP for customers in a multi-plant environment,” says SAP’s Said. “You can use SAP manufacturing integration intelligence to capture data around orders, batches and quality, and check to make sure the master data is accurate. It ensures process integration.”

The overriding strategy to assure the accuracy of plant data is to first assume that it’s inaccurate. Once you assume the data is corrupted, you can apply a number of programs, applications and models that can detect and fix data inaccuracies.

Sidebar Article - ERP System Ensures Accurate Data
To read the accompanying article to this story, go to
More in Control