AI-enabled quality control interface
2020
Context
EVRAZ is one of the largest steel and mining companies in Russia. The company operates across the full production cycle: from raw material extraction and steelmaking to the production of structural steel, rails, pipe products, and other materials used in infrastructure, industrial facilities, and large-scale engineering projects.
The project was connected to EVRAZ NTMK in Nizhny Tagil, one of the largest full-cycle metallurgical plants in Russia. In this type of production environment, even a small process improvement can have a significant economic impact. Production volumes are high, equipment runs continuously, and quality-control errors quickly turn into excess material use, defective products, and direct financial losses.
For EVRAZ, production digitalization was not an abstract innovation initiative, but a way to make quality control faster, more accurate, and less dependent on manual measurements. The company was implementing a system based on computer vision, laser measurement, and further AI analysis of production data. Our task was to design an interface that would help production teams see deviations in real time, understand the causes of problems, and make faster decisions about equipment adjustments.
-
Project Type
Industrial AI / Computer Vision / B2B / Manufacturing
-
Project Timeline
6 months
-
My Role
Senior Product Designer
-
Project Stack
Figma, Notion, Miro
Constraints and Metrics
The main goal of the project was to help production teams detect deviations in I-beam geometry faster and make decisions before an error turned into a defect.
The production environment created many constraints for the interface. A billet goes through several stages of deformation and turns from a 12-meter piece into an almost 100-meter beam. Everything happens quickly, in conditions of noise, steam, dust, high temperature, and poor visibility. Some workstations are located near hot metal, so the interface had to be readable from a distance and not require complex interaction.
Another layer of complexity came from the fact that different participants in the process had different tasks. Some users needed a quick signal that something was wrong, while others needed a detailed view of the entire beam length, rolling history, and the relationship between deviations and equipment settings. At the same time, employees could be located up to 300 meters apart and had to synchronize by radio, while quality data had previously been collected manually in separate Excel files.
We were not designing just a monitoring screen. We were designing the foundation for a system that could collect production data, analyze deviations, identify patterns, and later support AI-powered recommendations for equipment adjustment.
What is an I-beam?
An I-beam is a structural steel beam with an H-shaped cross-section. These beams are used in industrial buildings, bridges, warehouses, shopping centers, and other large structures where strength, stability, and efficient material use are critical.
At first glance, an I-beam may look like a simple product, but its production requires high precision. The cross-section, geometry, and weight must comply with GOST standards. If the beam is heavier than the standard allows, the company spends excess metal. If the parameters fall outside the allowed tolerances, the product may become defective.
That is why the key production challenge is to detect deviations as early as possible and adjust the process before material and production time are spent on a defective product.
Field research at the factory
The first stage was to understand the production process in detail. We needed to learn how I-beams were produced at that moment, where quality risks appeared, and at which point the interface could help the team make decisions faster.
We conducted more than 10 hours of workshops with the chief production technologist and built a CJM for both the current process and the future process after the implementation of computer vision.
In the current process, the billet is heated to a high temperature, goes through several stages of deformation, and stretches to almost 100 meters. During intermediate stages, workers visually assess quality, while precise measurements are performed only after the product cools down. If a deviation is discovered too late, part of the output may have to be written off as defective.
After analyzing the process, it became clear that the value of the interface was not simply to display new data on a screen. It had to help the production team quickly understand whether there was a deviation, how critical it was, where along the beam it occurred, what it could be related to, and what needed to be changed in the equipment settings.
After the workshops, we visited the production site in Nizhny Tagil to observe the process in real conditions and meet the future users of the system.
On site, it became clear that the interface could not be designed like a regular office dashboard. The shop floor was noisy, visibility was reduced by steam and dust, some workstations were located near hot metal, and a worker could spend up to 120 minutes per shift near the conveyor. Other workstations were located in dark rooms, where light interfaces created strong contrast and could increase eye strain.
This directly influenced the visual language of the interface. We had to think not only about data structure, but also about readability from a distance, contrast, speed of perception, minimal interaction, and how quickly a person could understand the state of the product in a complex production environment.
User roles
The initial hypothesis was that all participants in the process could use one shared interface. After interviews and observations, this hypothesis did not hold up. Each role had different tasks, different levels of responsibility, and different needs for detail.
The main research finding was that one universal interface could not support all users equally well. As a result, we divided the solution into several levels of detail depending on the role.
The rolling mill operator needed a simple information screen. This user controls the beam dimensions and needs to understand quickly whether there are deviations, without approaching the hot billet.
The senior rolling mill operator makes key decisions about the production process. This role needed a more detailed interface: beam length, cross-section, history, deviations, warnings, and the connection between a problem and equipment settings.
The equipment operator controls the equipment complex. This user needs to see data in the same context as the senior rolling mill operator, so the team can synchronize faster without unnecessary radio communication.
We also identified the shop manager as a separate role. This user does not need operational control near the mill, but rather an analytical dashboard to evaluate production metrics and economic impact.
Problems and solutions
real-time monitoring of geometry and weight
Before the implementation of computer vision, precise measurements were performed only after the product had cooled down. If a deviation was detected at that stage, the material had already been used, and the beam could become defective.
We designed an interface for monitoring size and weight that displays data immediately after the beam passes through the sensors.
For the rolling mill operator, the interface works as an information screen without complex interaction. The operator sees the beam cross-section, actual dimensions, deviations from the norm, beam length, defect zones, and borderline states. The allowed GOST values are shown nearby, so the worker can quickly compare the current state of the product with the standard.
The goal of this screen was not to make the user analyze tables, but to show immediately whether the product was within tolerance and where exactly the problem appeared.
red for defects and yellow for borderline conditions. Once the product passes through the sensors, the screen displays
a cross-section projection with its actual dimensions and deviations. GOST tolerance values help operators
quickly understand which deviations are critical.
role-based interface levels
The rolling mill operator needed a quick signal, while the senior rolling mill operator needed a detailed view of the full beam length, rolling history, and the impact of equipment settings.
We divided the interface by role and created a more detailed screen for the senior rolling mill operator. This user can analyze the entire beam length, select a specific section of the beam, view the corresponding cross-section, see the history of finished products, compare defective and acceptable output, monitor the rolling queue for the next few hours, and receive equipment status warnings.
We also added recalculation from hot to cold state. This is important because the product changes size after cooling, and the production team needs to understand in advance what the final weight and geometry of the beam will be.
This allowed the senior rolling mill operator not only to see the fact of a deviation, but also to understand the possible cause and the next step faster.
the cross-section view based on the selected area. Rolling history. Shows finished products for the selected section,
including defect ratio and length changes. Hot-to-cold switch. Shows recalculated
dimensions after cooling. System hints and warnings. Highlights critical conditions,
such as equipment operating close to maximum capacity.
3D and X-ray modes
If a defect repeats at a certain interval, it may indicate not a random error, but a problem with the equipment. In a numerical table, this kind of pattern is hard to notice quickly.
We added several visualization modes to turn technical data into understandable feedback.
The 3D View helps users see deviation patterns along the beam length. For example, if a deviation repeats at a certain interval, it may signal that the problem is connected to equipment behavior.
The X-ray mode shows recent rolled beams and helps the team quickly compare how additional adjustments affected the product dimensions. This approach helps the team not just record deviations, but see the dynamics of change and understand faster whether a correction worked.
The rolling history panel shows finished products for the selected section, including the ratio of defective to acceptable output and how the length changed over time. This helps the team evaluate production quality from the start of the shift and understand the result of the crew’s work.
The rolling queue shows the production schedule for the next three hours.
The interface also includes a hot-to-cold recalculation switch. This allows users to understand what the final weight and dimensions of the I-beam will be after the product cools down.
analytical dashboard for production efficiency
Quality and production data had been collected manually in separate Excel files. It was difficult for the shop manager to quickly understand the overall picture: where defects appeared, how weight changed, which products were at risk, and how this affected production economics.
We designed an analytics module for the shop manager. The dashboard allows users to move from the overall picture to details: first seeing results for a selected period, then reviewing deviations from target values, weight dynamics, products at risk, defect rate, and deviations in weight and size.
We also added a conversion of physical indicators into financial values, so the team could see not only technical deviations, but also the economic effect. This helped connect product quality with losses, missed value, and the real cost of production deviations.
This module also laid the foundation for further work with accumulated production data: pattern analysis, defect prediction, and automated recommendations for equipment adjustment.
Results
As a result of the project, we designed an interface for an I-beam quality control system that visualizes beam parameters in real time, helps teams detect deviations faster, shows defect zones and borderline states, supports different production roles, and helps connect deviations with possible equipment issues.
The solution reduces dependence on manual measurements and radio communication, while also giving the shop manager an analytical tool for evaluating quality and economic efficiency. An important part of the project was not only the visualization of current measurements, but also the foundation for further data analysis, defect prediction, and AI-powered recommendations.
The projected economic impact of the project was estimated at RUB 310 million per year through raw material savings and reduced losses from defects.
The system was implemented at EVRAZ NTMK in the wide-flange beam rolling shop. It combined laser measurement, computer vision, and production-data analysis to help operators detect deviations earlier and adjust the rolling process faster. The laser profiler automatically scans a 100-meter hot beam, measures its profile every 10 mm, and displays the data on the senior rolling mill operator’s monitor. According to public coverage of the implementation, after the system was launched, the shop saw several times fewer defects and faster work: employees could see product dimensions immediately and adjust rolling mill settings without additional manual measurements.
This project became one of my first examples of working with AI-enabled industrial systems.
The main lesson: in products like this, design cannot be limited to beautiful data visualization. The interface needs to help the user quickly understand what happened, how critical it is, why it may have happened, and what should be done next.
In industrial AI, trust, readability, speed of perception, and the right level of detail for different roles are especially important. The system may be technically complex, but for the user it has to become a clear working tool that helps them make decisions faster and with more confidence.