Monitoring of deflection of the M12 line

August 2024

#Predictive maintenance #Innovation #Working with students #Artificial intelligence #Special diagnostics #HW testing

Project's goal

The aim of the "PoC SU2025 AI Vibrodiagnostics" project is to use artificial intelligence to predict machinery failures based on vibrodiagnostic data in order to reduce equipment downtime, improve maintenance and increase operational reliability.

Results

The project "PoC SU2025 AI Vibrodiagnostics" presents an innovative approach to the use of artificial intelligence in the field of vibrodiagnostics to improve predictive maintenance, automation and monitoring of the condition of machinery. This project focuses on verifying the possibilities of using artificial intelligence for the evaluation of vibrodiagnostic data and their ability to predict machinery failures. If the prediction is successful, AI could effectively reduce downtime and increase the efficiency of maintenance planning.

Solution description

During the course of the project, a classification of failures on signals obtained from machinery in Škoda Auto was successfully carried out. This classification was performed using a 1D convolutional neural network and various approaches were investigated, including sequence and functional models, and multi-class and multi-label classification. The project shows that the best results for fault classification are obtained using accompanying parameters and a functional approach with multiple inputs.

One of the main benefits of this project is the possibility of reducing downtime thanks to the early detection of defects. If the prediction is also successful for combined failures and the solution is deployed on real data, maintenance could be significantly improved and equipment downtime minimized.

Another benefit is the finding that training artificial intelligence on simulation data is effective and allows models to be trained with appropriate accuracy. The ability of artificial intelligence to accurately identify fault types and vibration sources has also been confirmed, which is key to planning maintenance and reducing the risk of outages.

Overall, the "PoC SU2025 AI Vibrodiagnostics" project brings new perspectives in the field of vibrodiagnostics and artificial intelligence, and suggests possible improvements in the field of predictive maintenance, innovation and automation.

Benefits and Details

Reduced downtime – the project showed the potential for early fault detection and downtime prediction, which could significantly reduce equipment downtime and increase plant reliability.
Training AI simulation data – this hypothesis has been verified and it is clear that appropriately prepared simulation and test data is sufficient to train models with adequate accuracy, but it is necessary to train the models for combinations of failures.
Checking already processed defects and verifying whether the AI has found them – this hypothesis has been largely confirmed, in the case of individual failures, the detection of the type and source of failures was sufficiently accurate for the considered target expert system.
Evaluation of new data and their subsequent human control – this hypothesis has been confirmed, the output from the trained AI model for the blind test offers the worker a sufficiently specific and accurate indication of the type of failure and the specific bearing that is the primary source of vibration.