Email a friend
  • To include more than one recipient, please separate each email address with a semi-colon ';', to a maximum of 5

  • By submitting this article to a friend we reserve the right to contact them regarding Fastmarkets AMM subscriptions. Please ensure you have their consent before giving us their details.

Data drive

Apr 27, 2016 | 12:14 PM | Gregory DL Morris

Big Data, the ability to take huge volumes of information and assess them for relevant patterns and qualities, is not new to primary metals. Software vendors and service suppliers have been active in steel mills and aluminum smelters since the early 1990s and into the 2000s.

Since the global financial crisis, however, and the subsequent slump in demand for commodity metals, the tide has largely gone out. Several mills were the subject of white papers and case studies in data-driven manufacturing as recently as 2009 and 2010. But since then progress has been more isolated.

Top process control and enterprise management providers say they are working closely in a few cases with a plant here or there, and that research or pilot programs are under way at several metal makers. Also, sophisticated sensors and analytics are in use for predictive maintenance at many mills. But Big Data thought leaders say the metals sector is still in the very early days in making full use of the information it already has in hand.

“There has been a lot of discussion of true unstructured data correlation, especially for predictive maintenance, in the primary metals industry,” according to Yong The, a business consulting executive at Schneider Electric, “but so far we have not seen much actual uptake. That sector is still in the stage of classical condition monitoring. Metals companies are collecting data, but in the structured ways that they have been doing for years.” Schneider’s main manufacturing platform is Wonderware.

There has been a lot of research and development in metals, The said, and some of that has blossomed into operational applications. But those instances are only so far among the most “mature” of companies. “There are a few firms that are taking unstructured data from mining companies upstream, and analyzing that data to align their operations to meet the changing qualities of inputs. There are only a few companies doing this live, but there is a lot of R&D (research and development) in the space that is very promising,” The said.

Tim Sowell, chief architect for industrials at Schneider, said that “Big Data is being able to correlate patterns out of large amounts of unstructured data. Historically, operators have been logging hundreds of thousands of data points around any one asset as a way of predicting performance. But most people only look at the top two or three variables, at most the top 10 or 15,” over hours or days, at most weekly or month to month. That is structured data.

“Just in the past few years have people been trying to determine sets of events over the previous five or 10 years correlating across 100,000 points,” Sowell said. “That is unstructured. That is unleashing the data. The pattern recognition allows us to see things earlier than before, and possibly to identify patterns that would have been missed in structured analysis.”

In most steel plants, Sowell notes, “the historical data is the most underused asset. Operators look at the last 24 to 48 hours. But now everyone is trying to get the most out of their assets, and one of those assets is their data. Big Data takes the historical data to make predictive models.”

The said that in practical application around the industry, he is aware of a couple of leading-edge projects along those lines at copper smelters. “They are looking to optimize their concentrators by taking information from the mines. They are live, not pilot projects. They grew out of R&D efforts and are now in production.”

Sowell said that such ambitious data manipulation requires a high level of comfort and sophistication. “These are very mature operators in terms of their operational and technological development. They are among the world leaders in operational efficiency, and so it makes sense that they would be the first to try something like this.”

The concept of modeling in metals is not new, Sowell said. “I was doing it with casters in the 1980s. The difference is that now the mature companies are asking what more we can get out of our assets, and out of our technology. They are looking beyond assets to operational performance and productivity.”

A big part of the challenge in bringing Big Data capabilities to full implementation in the metals industry is the question of whether this is a business story that is moving out to the shop floor, or whether this is a technology story moving out from IT, or even a manufacturing and operations story moving in from the mill.

Stefan Koch, global leader for metals at SAP SE, said the answer is all of the above. “In primary metals you have already a great deal of data available. Some of that is hot data that goes to the dashboard to run the machine, but there is also data that needs to go to management for business reasons, and also data that needs to go to analysis for operations and efficiency, including predictive maintenance. To bring that all together with manufacturing is very powerful.”

He also notes that different processes within the metals chain have different needs in terms of predictive and analytical data. “In flat products there is a great deal of real-time influence from the data,” Koch said. “The values in the process affect the process and quality. In those cases the companies are looking for data of the moment for running the machine. Efficient operations and quality have to be maintained through the process.”

That is fine as far as it goes, but it does not go far enough, Koch said. “Operators in the caster may see conditions as within limits, but there could be some undetected operational or materials issue that could result in problems at the rolling mill. When something like that is detected at the rolling mill, it can be difficult to identify the cause, because each data set is in a separate silo.”

The idea, Koch said, is that at step three in the caster, Big Data can detect conditions that can be expected to lead to a given situation at step six. “If you know early on that you are at only 80 percent of a desired outcome, you can react earlier and make decisions earlier.”

SAP may be better known for higher-level enterprise-wide systems, but it has been among the software companies moving aggressively to fill the gap between management and operations. The company just released a new module for predictive maintenance and services (PdMS) on its Hana Cloud Platform. SAP has been a strong proponent of cloud-based services, but the PdMS module can also be run on site at a mill or foundry.

“There is definite value in coordinating business and operational data,” Koch noted. “That is the magic of predictive outcomes. It is not just a matter of whether the pump or valve needs repair, or if the slab or roll is of sufficient quality; the real question is whether the condition is better or worse for the business overall. Business and operations are very different in their culture, so they look at this data differently. That is the challenge.”

The next step after Big Data is Smart Data, Koch said. “As I mentioned, you have different types of data that have different relevance in time. That is the real value, not just an inventory of data to crunch, but tiers of data based on their relevance in time.”

Another goal with Smart Data is faster and more robust simulation, Koch said. “It used to take days or hours to answer the question, ‘What if...?’ With Smart Data alternatives can be evaluated very quickly and then implemented. You don’t need to do planning sessions overnight any more. You can run three alternatives on the fly.”

Another major control-systems company, Siemens AG, works through its XHQ Enterprise Operations Intelligence system. That is a platform for aggregating, relating and presenting operational and business data. Access to equipment health monitoring data can be made available to operations managers, equipment operators, sensor vendors and service personnel to provide important transparency for equipment health, and thus can be used to improve and sustain metals operations.

Those sensor vendors constitute an important component of the current state of data-driven manufacturing in metals. One major service provider is Woburn, Mass.-based Azima DLI Corp. “Metals were fairly early adopters of predictive maintenance,” according to Ken Piety, vice president of technology at Azima. “We worked with Alcoa (Inc.) in the early 1990s.” In 2006 the company published a white paper detailing the work it had done at Nucor Corp.’s Hickman, Ark., sheet mill. A Nucor executive said the initiative was considered successful but was discontinued several years ago.

“The start of the industry was a civilian adaptation of military technology, forward-looking infrared detection,” Piety said. “The technology was bulky and expensive. Service providers would drive up in big vans with heavy cables and guys with masters degrees would take measurements for a day. It was prohibitively expensive and complicated for a whole plant. Then in the ’80s we started to get microprocessors and PCs. Instrumentation came down in size and expense.”

The key, Piety said, is bringing the data to the people for interpretation, not the people out to the data. In addition to primary metals and other process industries, Azima is active in the nuclear industry and has the U.S. Navy as a client.

Based on about 30 years worth of work, Azima calculations show that for rotating equipment, maintenance and repair costs average about $19/horsepower (hp) for operators to fix things when they break. That cost can be cut by more than a third, to just $12/hp for scheduled, preventive maintenance. But that number can be cut to just $7/hp for predictive maintenance; that is 40 percent less than preventive maintenance, and just a third of the cost of waiting for things to break and then fixing them.

Obviously, the savings in time and labor is a significant portion of that, but Piety explains that for many complex pieces of equipment, the intrusion of an inspection or maintenance if it is scheduled but not really needed can do more harm than good.

Echoing the observation of the software firms that data-driven manufacturing can sometimes be branded as just a shop-floor initiative, Piety laments that predictive maintenance initiatives can be inconsistent. “Programs and champions within organizations come and go,” he said.

Even when programs are sustained, they can become victims of their own success, Piety explains. The major gains come at the start of the program as the most costly repairs are eliminated. After that the law of diminishing returns kicks in. Sometimes, when savings rates are not sustained, managers become disenchanted. But when the predictive maintenance ecosystem is dismantled, the costs revert.

Beyond predictive maintenance, data-driven manufacturing is also a powerful tool in benchmarking, according to Tyler Pietri, program manager at Azima. For some industries, such as oil and gas, Azima is able to benchmark individual plants or companies against other similar facilities or firms. But the numbers are not quite there yet in metals for direct peer-to-peer benchmarking, so metals firms are benchmarked against relevant similar industries. As more metals firms take up the idea, more closely matched benchmarking will be possible.

One of the side benefits of benchmarking is the sharp improvement in data compliance. “In one of our largest projects, data compliance went from about 50 percent to better than 95 percent in just two or three years,” Pietri said. That compliance can be regulatory or other government-mandated recording and reporting, and also internal industry or company standards.

Semeq Inc. is another global firm providing predictive maintenance assessments on rotating and electrical equipment. It is active with several steel mills, especially in Latin America, and also along the value chain through die casting, tube mills and simple fabrication to tier one suppliers to automakers. Standard measurements include vibrations, acoustics, oil parameters and thermography.

“We pull the data from the machines and take it to our lab for analysis,” Andrew Rodes, vice president and general manager at Semeq, said. “Vibration is the most complex analysis. There is a lot of math that has to be done to filter out the noise and determine real trends. But all techniques give early warning.”

Even as maintenance shifts from scheduled to predictive, readings are still mostly scheduled. “There is a standard curve for each different type of equipment that indicates the point in time from detection to failure,” Rodes said. “For vibration we need two measurements within 60 days, so we gather data from 95 percent of our customers every month.” Oil is sampled every 90 days.

Other techniques need sampling less often, quarterly, semi-annual, or annual. There is a “bathtub curve” for maintenance costs as predictive maintenance takes effect: there is a steep drop at first as major issues are addressed. That then flattens as a new normal is set—lower costs, higher efficiency and reliability. Then finally an increase in costs again as components near the end of their service life.

“There is also a parameter for intensity of use. “In steel mills the equipment is running all of the time,” Rodes said. “That has a different use pattern than a specialty line that might be run only every now and then.”

While there is some work in the area of remote sensing, Rodes notes two important factors that militate against it. One is the electromagnetic reality that many parts of the plant are Faraday cages where signals cannot get through. The other is the value of having human hands and eyes and ears on the shop floor.

“Every maintenance department is understaffed,” Rodes said. “I have never seen a maintenance staff anywhere that has enough people, enough time and enough budget. And some plants are so big that the staff may not even see a piece of equipment, they may never walk past it. So it helps to have our people out in the plant.”

Global Shop Solutions, The Woodlands, Texas, is an enterprise resource planning (ERP) supplier that has just added a new module called True View specifically to represent the shop floor. “The work center has a dispatch list for each machine,” Michael Melzer, vice president of operations, said. “That ensures that the materials and equipment for each job are part of the work order. It’s all on the dashboard, ready to run.”

The idea of a work center dashboard is big in the auto industry, Melzer noted. There is a mobile system for materials handling that can be accessed from hand-held devices. That helps tag parts as good or scrap. That includes materials properties, including chemistry, metallurgy, tensile and yield strength, among others.

“Batch sampling is increasing in many industries downstream from primary metals,” Melzer said, “including aerospace and energy. That helps manufacturing and purchasing plan for longer-term operations. Our customers in metal are moving from management by fire, in which they respond to orders, to forecasting and anticipating. That means they are able to win more business and repeat business.”


Latest Pricing Trends Year Over Year