EN
PL
Challenge
Polteknik Sp. z o.o. specializes in supplying machines and technologies for metal processing, as well as in the robotization of production workstations and automation of production lines. One of the key elements of the company’s offering is sheet metal bending machines, which often constitute a critical component of production lines in many industries, such as automotive, home appliances (white goods), and the metal industry.
Any unplanned downtime of such machines generates very high costs – both direct (loss of production, service costs) and indirect (delays in order fulfillment, contractual penalties, reduced quality).
The objective of the project was to develop an IT solution that would enable the detection of early symptoms of increasing component wear or elevated risk of machine failure, allowing preventive actions to be taken sufficiently early. In practice, this meant transforming the maintenance model from reactive to predictive. An additional requirement was the scalability of the solution and the ability to deploy it across many machines and installations worldwide.
Solution
The project was carried out through close cooperation between three companies: Polteknik Sp. z o.o., StatSoft Polska Sp. z o.o., and STIGO Sp. z o.o.
Polteknik Sp. z o.o. carried out activities related to the acquisition of data directly from machines. The scope of work included the identification of key measurement points in the sheet metal bending process. Subsequently, an experienced service team selected and installed sensors to monitor machine operating parameters. The sensor data were integrated with a data analysis system, enabling observed deviations in the data to be linked to the actual technical condition of the machines.
As part of the project, Polteknik:
-
ensured continuous and structured access to process data,
-
supported the analytical team in data interpretation,
-
provided expertise on process flow, characteristic machine operating modes, and typical component wear mechanisms.
StatSoft Polska was responsible for the entire analytical process – from data preparation and cleaning, through building and selecting the best predictive models, to their implementation in a specially designed database.
Based on data collected from multiple machines, an early warning system was developed using advanced machine learning methods. An unsupervised approach was applied, leveraging PCA, KNN, MCD, and autoencoder algorithms, which learn the typical behavior of a machine and identify deviations from the norm. A key outcome of the work was the determination of an anomaly score -a synthetic indicator of failure risk level. Additionally, supervised models were considered and tested to forecast future values of the anomaly score, enabling even earlier signaling of increasing risk of irregularities.
The system architecture and database were designed in a flexible and scalable manner, allowing easy adaptation of the solution to new machines, sensors, and additional data sources, while also ensuring readiness for further development.
Based on the developed model, STIGO prepared software with an intuitive user interface, enabling:
- easy access to analysis results and continuous monitoring of machine condition,
- automatic acquisition of machine data and operating parameters,
- execution of forecasts and generation of warning signals.
Results
Thanks to the implementation of the solution:
- machine users gained a tool supporting informed and optimal planning of technical inspections,
- it became possible to replace wearing components before failures occurred,
- the risk of unplanned production downtime was significantly reduced,
- maintenance teams gained better real-time visibility into the technical condition of machines,
- the machine manufacturer strengthened its competitive advantage by offering customers solutions that increase reliability, availability, and operational safety of equipment.
Summary
The project demonstrates how advanced data analytics and machine learning can effectively support industry in the transition from reactive to predictive maintenance. In an industry where every hour of downtime translates into measurable financial losses, such solutions are no longer an add-on but a key element of modern production systems.
Implementations of this type not only increase machine reliability but also build long-term value—both for equipment manufacturers and their end customers, who expect stability, predictability, and maximum process efficiency.