- By Sheetal Birla & Chris Lee
- November 09, 2020
While projections for steel remain positive in the near future, growth has been slowing and prices have been under downward pressure since early 2018. How does one meet increasing demand in such an uncertain environment? IIoT and Industry 4.0 technologies can provide an answer: next-generation reliability for the steel industry.
The steel manufacturing sector, like many others, is facing uncertain times exacerbated by the COVID-19 pandemic’s effect on the general economic demand. While projections for steel remain positive in the near future, growth has been slowing and prices have been under downward pressure since early 2018. How does one meet increasing demand in such an uncertain environment while minimizing the risk of over investment? Based on our experience working with a major steel manufacturer, we believe that IIoT and Industry 4.0 technologies provide an answer.
In the past, options for increasing output were limited. Typically, foundries and mills would procure additional equipment in order to boost production. This approach has significant financial (capital investment), technical (procuring, commissioning, characterizing and ramping new technology) and schedule risks. Industry 4.0 and IIoT provides a solution to this problem which industry leaders across several industries have successfully deployed to increase their competitive advantage. Because of these risks and the new options, a capex heavy plan may not be the best approach.
Enter Industry 4.0 and an AI based approach
Advanced analytics, AI, big data, smart sensors, enhanced connectivity and the vast computing resources of the cloud open the way to reclaim lost productivity from existing assets. Such technologies, especially software based ones, can be deployed at a fraction of new equipment cost.
Two important metrics here are availability and quality. If an asset is unavailable, it is not running and, therefore, not producing. Likewise, if the output has too many defects, the output cannot be sold. One customer estimates losses of thousands of tons per year due to unplanned downtime alone. By using Industry 4.0 technologies, the difficult-to-predict, adverse equipment behaviors can be understood and managed to reduce process and manufacturing disruptions. In particular, predictive analytics solutions which leverage AI in order to detect patterns of equipment operation are well suited to addressing this problem. AI brings new ways to discover important failure modes and to detect important conditions which indicate that a service cycle is needed. Such predictive analytic capabilities supplement existing Asset Performance Management (APM) strategies and help to create an effective condition based maintenance (CBM) plan.
While non-AI based methods, like rule based triggers and regression analysis, can be used to define and detect conditions requiring maintenance, they have several limitations. They may not capture the complex, multivariate nature of the problems or they are effective only for smooth changes over time, which, as our experience with one of our steel customers has shown, is not always the case. We have found that a pattern based approach is more effective. In this approach, AI is used to detect behaviors of equipment behavior and to correlate those with a particular failure type. Patterns greatly reduce the difficulty of implementing condition monitoring for the reliability engineer while increasing the sensitivity to a range of failure modes.
There are three basic ways to use AI-driven pattern based monitoring in condition based maintenance:
1. Discovering conditions for maintenance: Patterns discovered in the operational data can help reliability experts find failure modes without requiring the expert to explicitly define the behavior of interest. Rather than creating a set of rules for a potential failure (e.g. when pressure is greater than X and temperature is greater than Y and…), they can just label examples of the potential failures in the data. Using AI in this way allows detection of conditions which may be too difficult or subtle for a human to define.
2. Performing condition monitoring: It is not enough to define the conditions under which maintenance should occur; it is necessary to use real time data to determine if the assets have met those trigger conditions. In simpler times, this may have been done using conditional formulae in spreadsheets or trend charts in BI tools but these are not up to the task for a pattern-based approach. AI is critical in condition monitoring since translating pattern detection algorithms into a spreadsheet or a line chart and making deductions from it is beyond the scope of human capabilities.
3. Understanding condition based alerts: Before creating a work order, it is helpful to check data in the period of potential failure for information which helps confirm root cause. For example, by reviewing which sensors were most involved in a warning, the engineer can see whether the pattern is electrical or hydraulic and prepare the service team accordingly. Good AI solutions can provide explanatory information which helps those teams compare the current condition to known reference conditions and verify what has actually occurred.
Useful as it is, AI can be difficult to set up and maintain. Most AI solutions are built for data scientists who possess specialized knowledge in data preparation and algorithm tuning. A CBM plan which requires the regular input of such specialists cannot scale up to the degree needed to solve the business problem. Enter “Operational AI,” an easy-to-use AI built for operational teams to overcome these barriers. Operational AI solutions make pattern based condition monitoring a reality for most mills and foundries.
Examples of applied Operational AI in steel manufacturing
Following are two examples where Falkonry’s Operational AI was used in a steel casting line. These are taken from a comprehensive set of cases covering both continuous casting and rolling.
Warning patterns in segments operational data predict early failure
After the thin slab comes out of the caster’s mold it passes through the segments. The 3 segments in the vertical caster guide the steel strand, adjust the thickness while maintaining precise shape and cooling it as it passes it onto the pinch rollers. Due to heat, friction and stress, the rollers in the segments will gradually wear out, go out of alignment or get stuck leading to wedge & camber problems and down time. The segments are replaced based on either length of the cast strand or on a fixed schedule. Estimating actual roller wear or detecting gradual misalignment is a difficult process. Unexpected failure can stop the entire production line while too frequent service reduces the line’s output.
Operational data from 150 sensors, including periods of both normal operation and segments roller failure events was loaded into Falkonry. Signals from the data historian included segments’ roller forces, positions, offsets, calibrations, rod & piston pressures, slab width, thickness, steel types, and casting speeds.
Falkonry’s Operational AI software was used to observe patterns over a few months of the segments rollers’ operation. Using these insights, combined with known failure events, the users labeled periods of normal operation and warning periods prior to failures such as segments roll alignment deviations, rollers getting stuck, and high force events. Falkonry software used these examples to identify ‘normal’ and ‘warning’ patterns. This learning was applied to the remaining data, resulting in the users being alerted to warning patterns 7-10 days in advance of actual failure.
The predictions were then reviewed using Falkonry’s explanation scores in order to understand which signals were most involved in the problem. This information was highly valuable to the operations team, ensuring that proper corrective maintenance was performed.
The Operational AI was then deployed via Falkonry Analyzers to collect and classify live data from the caster.
Warning patterns in shears operational data predict early failure Pendulum shears cut the continuously cast steel strip into predetermined lengths. The pendulum shears motor and hydraulic system go through high operational stress variations. Unexpected shears failures cause the casting line to stop, leading to down time and production losses.
Operational data from 14 sensors, including periods of both normal operation and shears failure events was loaded into Falkonry. Signals from the data historian included hydraulic pressures, motor current, piston positions, temperatures, and casting speeds.
Unsupervised learning in Falkonry’s Operational AI software identified important patterns of behavior, giving operational experts a better idea of normal and problem periods of operation. Insight from these patterns was used to label periods of warning behavior preceding failures. This learning was applied to the remaining data to confirm that events of interest could be predicted using the precursor patterns identified. Falkonry discovered warning patterns 2 weeks in advance of shears issues and failures.
Lessons for success
From our experience of deploying Operational AI for clients in steel and other industries, we have found certain approaches to be more effective at overcoming the barriers that typically hinder adoption.
Following are some of the key lessons we have learned:
Proof of concept projects do not affect real production outcomes and therefore tend to be ignored. Even if they do show good results, those results rarely affect today’s operations making it difficult to justify a purchase. Instead, it is more effective to go straight to a limited pilot engagement where the operations team uses the Operational AI solution as a part of their daily work and status meetings. In this way, real problems are solved and the impact on asset performance is visceral and clear. The question shifts from: “why should I turn this on?” to “what do I lose when this is turned off?”
Operational AI is easy enough for existing teams of operations experts to use. By putting the ability to monitor conditions in the hands of the people who best understand what to look for and which assets should be looked at, important problems are solved more quickly. Learning is distributed instead of gated through a data science expert. In one case, we saw how the agility provided by operational AI accelerated a performance management strategy decision by several months, shaving valuable time off of a cycle that was typically years long
Technology to make better decisions is not enough if the information doesn’t get to the people who need to act on it. Selecting an Operational AI solution which integrates well with existing business processes and support software (like EAM and CMMS systems) and working with the end-users is important to realizing the promise of better CBM.
To help steel manufacturers succeed in this uncertain business environment, Industry 4.0 and IIoT technologies provide lower cost, lower risk approaches to increase foundry and mill output by extracting the most availability and quality from existing equipment. Achieving these goals through better condition based maintenance is possible with the advances in AI and modern cloud computing. However, because traditional AI is difficult to set up and use, it does not scale fast enough to provide the desired ROI. Operational AI was created to fill this gap. Deploying operational AI, steel manufacturers can find conditions which result in unplanned downtime allowing them to optimize maintenance, increase availability and achieve their increased production targets.Read more
Did you enjoy this great article?
Check out our free e-newsletters to read more great articles..Subscribe