PNIPU: ACS by the process of polishing turbine blades

The main articles are:

• APCS - typical structure
• Video analytics (terms, applications, technologies)

2025: Creation of ACS by the process of polishing turbine blades

Scientists of the Perm Polytechnic University have created a smart flaw detection of turbine blades Blades are key elements of the design of turbines in aviation and. power industries They convert the energy flow into mechanical rotation of the motor shaft. Their quality and reliability are extremely high. The university announced this on September 10, 2025.

𝓲

Фото: PNIPU

Of particular importance in the manufacture of blades is the process of flaw detection after their processing. Defects are detected manually, which requires high concentration, a lot of time and is complicated by the non-standard shape of the part. Scientists of the Perm Polytechnic University have developed an automated system for controlling the process of polishing turbine blades using intelligent video analytics. This technology allows you to monitor the quality of surface treatment in real time, detect defects with 96% accuracy and automatically adjust the process without human input.

Turbine blades have complex geometrical shape with aerodynamic surface. As of September 2025, their polishing takes place on special machines or using a robotic manipulator. Part is installed into device and basic trajectory of polishing tool motion is set through program. But after processing, the controller operator must visually carefully examine the entire surface for scratches and various defective traces. It is extremely difficult and inconvenient for a person to examine it from all angles. This takes a long time and in the event of an oversight, there is a risk of a part with a defect entering the next stage of production.

Scientists have developed a comprehensive neural network model with video analytics that combines these processes into one stage - the system combines blade processing and defect control in one automated cycle. This greatly improves the accuracy and speed of production of key components of aircraft and industrial engines.

At the heart of our software is artificial intelligence, capable of recognizing various types of defects. The system is a complex of hardware and software solutions: a special video camera is fixed on the hand of an industrial robotic manipulator performing polishing. Along a pre-calculated trajectory, its movement and inspection of the blade takes place from all the necessary angles, even in hard-to-reach places. Then, in real time, a powerful computing complex processes the video stream using a trained neural network. All the slightest anomalies - scratches, chips, irregularities of polishing - are recorded, and after the scan is completed, a detailed report is generated for the operator. In turn, he can launch additional polishing of exactly those areas where it is necessary to eliminate the flaw, - explained Daniil Kurushin, Associate Professor of the Department of Information Technologies and Automated Systems of PNIPU, Candidate of Technical Sciences.

To analyze the video stream, scientists have chosen and adapted one of the most modern architectures of neural networks - YOLO11. Her training was carried out on an extensive database - more than 1,500 images of blades of different shapes and with various types of defects, taken at several angles and in conditions of special ultraviolet illumination. A prototype program interface has already been developed. After fixing the blade in the working area, the operator only needs to select the type of blade in the system - he already has his own mathematical model and program of robot movements. When started, a field with found defective objects and a full report on the analysis is displayed in a convenient form.

The survey of reference defects for training the neural network and testing of the prototype were carried out directly at the production site of the engine-building plant. This ensured high practical significance and compliance of the system with real production tasks. The recognition accuracy was 96%, which indicates the high ability of the model to correctly classify the polishing state. And completeness - 94%, which indicates the ability to identify most defects, - said Alexey Dukhanin, graduate student of the Department of Information Technologies and Automated Systems of PNIPU.

An intelligent system based on deep machine learning and computer vision optimizes the process of polishing turbine blades based on their geometry and material properties. The technology makes it possible to detect microscopic surface defects in real time, which significantly improves the quality and efficiency of production of critical components. As of September 2025, discussions are underway on the introduction of technology at one of the industrial enterprises.

In the future, the development team plans to expand the database to improve recognition accuracy, add 3D blade models to create more informative reports, as well as scale the architecture of the system, expand its application and integrate it with new advanced technologies. This will optimize production processes and increase competitiveness in the market.

The introduction of the technology of scientists of the Perm Polytechnic University at enterprises of aviation and energy engineering will increase the quality of products due to 100% automated control and exclusion of the human factor, reduce the control time compared to traditional methods of visual inspection and reduce the cost of scrap and refinement of products.