“It is a mix of computer scientists and mathematicians with years of experience. We can connect to existing systems within companies and design ways to make better use of their data. We also recommend which AI tools to add to make the entire corporate ecosystem even smarter,” says Hynek Cihlář, Aricoma's AI solutions architect and AI consultant, enthusiastically.
Listening to him, one gets the sense that we are living in a remarkable time when extraordinary things are becoming possible.
It's not a tool, but a strategy
Hynek Cihlář and his colleagues are currently implementing artificial intelligence tools on a large scale, mainly in manufacturing processes. As he himself says, it is not so much a matter of adding a single tool to the production line or conveyor belt. That is only the final stage. The magic of it all (and what leads to an increase in product processing speed of up to 50 percent) lies in the design of the entire ecosystem.
“We receive requests from both smaller and larger companies. Most of them already have some kind of ERP (production and related process planning) or MES (real-time production monitoring) system, but they know that they are not using them to their full potential. Or they don't know how to work with all the data that the systems provide. That's what we propose to them,” describes Hynek Cihlář.
According to him, a typical request looks like this: company managers either know about a specific problem that bothers them when handling their own production and data. Or, on the contrary, they vaguely suspect that artificial intelligence could help them work more efficiently, but they have no idea how to go about it. “It's important to note that we don't usually revolutionize things. More often than not, we take an agile approach and build on top of existing systems. Starting with one workshop or production line, and then moving on.”
An extra partner
One of the advantages of today's large language (so-called multimodal) models is their ability to generate text descriptions of visual input. If the system is properly trained, it behaves like a partner with its own thinking. For example, a camera-based quality monitoring system can be enhanced to mark defective areas automatically. The image is then sent to a multimodal language model, which, based on the context in the service documentation, not only describes what it sees, but also makes a logical interpretation, guiding the employee in question on how to solve the problem.
A good example of this is a project for Vars Brno, a company that monitors the condition of Czech road infrastructure. “Their measuring vehicle uses a radio beam and a camera to examine how damaged the road surface is. We trained a neural network for them that recognizes cracks and the extent of road damage, and maintenance is now planned accordingly. We shortened the entire process from several weeks to about 60 hours,” says Hynek Cihlář, giving one of many examples of its use. You can read more examples alongside this article.
Where in industry and logistics can artificial intelligence tools be useful?
1. Predictive analysis of production cycle quality
Product quality issues are often only discovered at the end of the production cycle, leading to high repair costs or complete disposal. AI tools that continuously analyse production sensor data, visual process data, and historical data on the quality of finished products can identify correlations and patterns that indicate potential quality issues before they occur. For example, they can detect minor changes in material texture and, based on the records, predict the risk of breakage. The system generates timely alerts for operators, reducing the costs caused by poor-quality production by 15 to 25 percent.
2. Dynamic optimization of production parameters
Production parameter adjustments are often static or made manually based on experience, which reduces efficiency. AI monitors machine sensor data, visual data from the production process, energy consumption information, and text records of manual adjustments, enabling it to recommend optimal settings for each stage of the production process to maximize efficiency and minimize waste. The system continuously learns and adapts. The result is a 5 to 10 percent increase in production efficiency, along with reduced energy and material consumption.
3. Predictive maintenance of production machines and equipment
Machine failures cause unplanned downtime and high costs. Scheduled maintenance is often ineffective because it is performed unnecessarily early or late. AI, on the other hand, analyses data from machine sensors, visual data (e.g., wear and tear on parts from cameras), text records (maintenance history, manuals), and audio recordings (unusual machine noises). This enables it to predict impending failures with high accuracy and recommend the optimal time for maintenance. The result is a 20 to 40 percent reduction in unplanned downtime and extended machine life.
4. Automated visual inspection with advanced defect detection
Manual visual inspection is time-consuming, error-prone, and subjective. Traditional machine vision may struggle to detect complex or unexpected defects. However, modern machine vision processes high-resolution image and video data from the production line. It can identify correlations in images, such as microscopic defects, deviations in colour, shape, structure, or surface, and convert them into text descriptions, even for highly variable products such as natural materials. This improves defect interpretation and categorization by 20 to 30 percent and increases inspection speed by up to 50 percent.
5. Optimizing warehouse layout and real-time package tracking
Inefficient warehouse layout leads to long picking times and difficulty locating goods. Manual shipment tracking is prone to errors and delays. By integrating with IoT sensors and cameras, artificial intelligence can identify the position of each item and shipment, compare it to the plan, and alert you to deviations. It can also visually identify damaged goods upon receipt or shipment. This reduces picking time by 20 to 30 percent.
6. Predictive inventory management and automated ordering
Imbalanced inventory levels lead either to excessive storage costs or to material shortages and production interruptions. This can be prevented by analysing historical sales data, seasonal trends, macroeconomic indicators, and, if necessary, image data from warehouse monitoring. Based on this data, AI predicts future demand with greater accuracy and automatically generates material order proposals for suppliers. This reduces storage costs by 10 to 15 percent and also helps cash flow.