AI and Machine Learning in Steel Manufacturing:

What Industry’s AI Transformation Means to Steel-Producing Companies

In this article, we will investigate what AI means for the steel industry, look into the pros, cons, and challenges of AI adoption, and explore major use cases that steel manufacturers can instrumentalize and take home right now.

Steelmaking is an advanced and complex industry in which sophisticated processes of steel manufacturing are dynamically interlinked with production chain operations and workflows, and the supply chain. Historically, only the most competitive and sustainable companies in the field have been able to get their slice of the pie on this market. And the pie is worth fighting for.

According to Oxford Economics and WorldSteel Association, in 2017 the steel industry sold $2.5 trillion worth of products and created $500 billion value added. Slabs, hot-rolled coil, cold-rolled coil, rebars, and rails are definitely a good business.

As more industries get transformed and disrupted by digital technologies, the steel industry is perfectly positioned to overcome challenges of AI transformation — the technology, the funds, and the applicable use cases are there.

AI in Steel Industry: Overview

Steel is an extremely data-intensive field. Every steel manufacturing process generates dozens of thousands of data points. And data is the fuel for AI and machine learning.

 

Here’s how AI & ML use data, in layman terms:

  1. Based on a predefined problem, an appropriate use case is selected

  2. Data collection and data review are initiated, to meet use case criteria

  3. ML algorithm(s) is designed and built (based on the collected data)

  4. Available data is split into two major pools, for training and for testing 

  5. Training data is used to train and tune the algorithm, to ensure best performance and accuracy

  6. The algorithm’s performance and accuracy are verified using testing data

  7. The algorithm is deployed in production in case of positive testing; otherwise, it gets re-trained and re-tuned

This framework works only if the organization pursuing the AI transformation has the right data for its use case; i.e. they should have a hefty amount of high-quality, structured or unstructured data in their data collection and processing system. And the steel industry is hardly the exception.

Let’s say that you collect data from hot metal detectors and pressure sensors but want to develop an AI-powered defect detection system for your steel products. You might not have the right type of data in place — namely, data from laser scanners, or visual data captured by cameras, containing various defects. In this case, you will have to collect data first, and only then start exploring the design options of the AI solution. Because in AI, data is king.

Challenges of AI Adoption in Steel Industry

Aside from data, other challenges exist as well, though.

 

Steelmaking processes are extremely complex operations-wise, and they are multi-physics processes, too. Input variables change all the time, and they highly depend on correlations between themselves and ever-changing environmental conditions. In practical terms, this means that AI professionals, ML engineers, and data scientists may lack the knowledge and experience required to come up with a reasonable solution — this necessitates tight cooperation with process operators. Manufacturers should be ready for knowledge transfer from their employees to engineers.

Another challenge is integration of AI/ML solutions. They can be built as part of the factory’s existing on-premises data collection and processing system, hosted entirely in the cloud (e.g. AWS, GCP, Azure), or split between on-premises and the cloud as a hybrid system. The solution design depends on the type of data collected, use case, factory’s technology environment, and more. The trend is, however, to move away from on-premises to the cloud.

 

Overall, data quality and quantity, selecting an appropriate use case, finding the right talent, and in-house operational and technology silos are major obstacles to AI adoption in pretty much every industry.

Benefits of Implementing AI Solutions in Steelmaking

AI and machine learning have huge potential in the steel industry. Let’s take a closer look at some of the benefits they offer.

  • Automation of routine tasks. Around 80% of predictable and 25% of unpredictable  physical work activities in manufacturing can be automated, according to McKinsey. And though steel is less susceptible to automation than most industries on average, the shift to “smart” human-computer systems is projected to increase performance, operational safety, and worker satisfaction.

  • Improved worker safety. Steelmaking is dangerous. To stay safe in the steel mill, workers need to strictly follow safety rules and regulations, wear personal protective equipment (PPE), and reduce the amount of manual handling and repetitive work. AI can enforce PPE compliance, monitor and track safety, and automate physically straining routine tasks that generally cause injuries due to long hours of handling, lifting, and carrying.

  • Enhancing process models. Steel manufacturing processes are wired for quality, efficiency, and sustainability. It is in every manufacturer’s interest to ensure ease of operation, speed up production cycles, reduce material consumption, improve health of the equipment, and lower downtime. AI complements and enhances classical process models by allowing them to dynamically evolve based on input data, thus establishing fast and resilient feedback loops.

  • Production cycle improvements. More sophisticated, dynamic processes ensure incremental increases in performance, productivity, and product quality. More efficiencies can be achieved through AI’s data analysis and analytics capabilities. AI can optimize material movement in and product movement out of the steep mill, assess material consumption, and predict equipment maintenance time, thus ensuring overall efficiency of the production and supply chains.

  • Increased product quality. AI and machine learning systems play a vital role in quality control. Through automation, steel manufacturers can ensure that defects are detected faster and at scale. “Smart” defect detectors find errors that humans cannot realistically see by using high-resolution cameras and multiple IoT sensors. On top of that, the detectors can be fused into production at much earlier stages, to guarantee that defects are detected when the damage is mostly recuperable.

Below is the example of how deep learning-powered defect detector “sees” different types of defect on different surfaces (from Steel Surface Defect Diagnostics Using Deep Convolutional Neural Network and Class Activation Map):

AI and machine learning allow steel manufacturers to automate more parts of the production process. They used to have programmable computers at their disposal. Now, however, they are equipped with self-learning systems that process and analyze to find patterns in complex data, predict events based on these patterns, and create human-to-machine and machine-to-machine feedback loops. All of that results in lower production and operational costs, better product quality and efficiency, and higher revenue.

Use Cases of AI in Steel Industry

A wide range of technological breakthroughs, from Big Data and data lakes to the cloud and high-performance computing to 5G and industrial IoT, enable multiple applications of artificial intelligence and machine learning in the steel industry.

#1 Predictive Analytics

Production errors (often caused by human factors) are extremely costly in steelmaking. Scrap and rework take up inventory space, force re-inspections and rescheduling, and reduce steel mill’s overall production capacity.

 

To streamline production processes, workers should have the tools to estimate and predict “error potential” at different stages of production. 

For example, let’s look at clogging, a prediction error that causes massive downtime. When the steel casting process is in progress, workers cannot assess and measure the clogging level in the nozzle; they have to change the nozzle after the clogging event, not in advance.

Thank you for reading! 

Feel free to rate the article:

The predictive analytics ability of AI allows the operators to assess various data points, including heat levels, vacuum degassing measurements, casting speed, rod position, etc. to predict the clogging event with high accuracy. In doing so, workers supply a new nozzle in advance, which reduces downtime.

 

Other notable cases mainly relate to continuous casting (e.g. degassing process prediction, steel temperature analytics, defect prediction based on fluctuating continuous casting parameters) and process management (e.g. material movement & handling, process completion time estimation, demand forecasting).

#2 Predictive Maintenance

 

AI and machine learning enable steel manufacturers to implement predictive, not reactive or preventive maintenance of equipment. This results in:

  • Faster, proactive fault detection

  • Improved production efficiency

  • Lower maintenance costs

  • Increased equipment life

  • Reduced equipment outages & downtime

 

Having predictive vs. preventive maintenance systems in place proves critical, given the cost of industrial machinery downtime in primary metal industries — $44,000 per hour

#3 Defect Detection & Quality Assurance

In steelmaking, scrap metal is recycled and reworked. The job of the manufacturer is to minimize the amounts of scrap and rework by implementing comprehensive defect detection and rigorous quality assurance methods. 

 

In practice, this often means several things:

  • Moving away from manual to automated defect detection

  • Hiring more QA professionals to monitor quality

 

Human quality checkers, however, detect defects at later stages of production — oftentimes when the product is ready to be shipped — which means an extremely high cost of rework. In the meantime, traditional defect detection systems based on automated optical inspection (AOI) are clumsy, nuanced, and costly to implement.

With the rapid transition towards Industry 4.0., defect detection systems powered by machine learning and deep learning present a viable alternative. Such systems take advantage of industrial IoT and high-resolution cameras that feed cloud-hosted algorithms with various data points to self-learn and identify specific “error” patterns (i.e. anomalies), not just match predefined patterns and templates against captured data as AOI systems do. 

 

Below are different types of defects detected by a DL-powered system on the surface of a hot-rolled steel sheet: a) crazing, b) folding, c) inclusion, d) original, e) patch, f) pitted surface, g) rolled-in scale, h) scratch.

The steel defect detection solution by VITech Lab operates in a similar fashion, by capturing visual data by high-frequency cameras, processing it in the cloud, and pushing alerts to QA systems and quality engineers, in real-time. The solution detects and classifies defects with 99% precision.

#4 Real-Time Production Monitoring & Tracking

Real-time monitoring & tracking and near real-time reacting on specific activities is one of the most controversial applications of AI in general. It raises the issues of privacy, safety, and trust. Many are worried such AI systems are too inaccurate, and that they can be biased.

 

In workplace environment, however, monitoring and tracking systems are seen as an advancement in workplace security and employee convenience. For example, you can use them to ID workers for access, or check if they comply with PPE guidelines. 

 

In the steel industry, such systems can also be used for worker safety, yet it makes more sense to monitor and track how industrial machinery operates, and how materials get moved across the factory floor. This allows to optimize supply chain and inventory and to reduce downtime.

Conclusion

AI and machine learning are posed to change the business and technology landscape in the 21st century. Spurred by the booming growth in data collection and processing capabilities, AI allows organizations to bring process automation and decision-making on an entirely new level. We are quickly progressing from programmable to self-learning “smart” machines.

 

AI presents a huge opportunity for steel manufacturers. They have the time and resources needed to drive comprehensive AI transformations, to increase efficiency, improve product quality, reduce operational costs, and, ultimately, generate more revenue.

#4 Real-Time Production Monitoring & Tracking

Real-time monitoring & tracking and near real-time reacting on specific activities is one of the most controversial applications of AI in general. It raises the issues of privacy, safety, and trust. Many are worried such AI systems are too inaccurate, and that they can be biased.

 

In workplace environment, however, monitoring and tracking systems are seen as an advancement in workplace security and employee convenience. For example, you can use them to ID workers for access, or check if they comply with PPE guidelines. 

 

In the steel industry, such systems can also be used for worker safety, yet it makes more sense to monitor and track how industrial machinery operates, and how materials get moved across the factory floor. This allows to optimize supply chain and inventory and to reduce downtime.