Bearing Defect Prediction Using Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting the evolution of bearing defects rely on manual visual inspections, which are time-consuming, prone to expert interpretation errors, and require expert presence, leading to inconsistent defect analysis and delayed maintenance planning, especially as they fail to account for surface features like oil residues and lighting reflections.
Innovation Solution
A method utilizing a trained deep learning algorithm, specifically a neuronal network, to identify defects and predict their evolution from pictures of bearings, extracting geometrical parameters and considering operating parameters and bearing models, enabling automated defect classification and maintenance recommendations without expert intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual visual inspection by experts is used to identify and predict bearing defects, then defect analysis can be performed with human expertise, but the process is time-consuming and requires expert presence on site
Solution Approach 1:
The patent replaces the manual visual inspection process with an automated image processing system that uses computer algorithms to analyze bearing pictures. The system extracts geometrical parameters of defects automatically and predicts defect evolution using mathematical models, eliminating the need for expert presence while maintaining accurate defect analysis.
Solution Approach 2:
The system creates a digital representation of the bearing defect by taking pictures and processing them through image analysis algorithms. The geometrical parameters extracted from these digital copies (area, perimeter, shape factors) are used to predict defect evolution, replacing the need for physical inspection by experts.
2Reliability
If expert interpretation of bearing pictures is used, then defect evolution can be predicted based on expert knowledge, but the results are inconsistent when multiple experts interpret the same pictures
Solution Approach 1:
The patent transforms the subjective expert interpretation process into an objective parameter-based system. Instead of relying on expert opinion, the system extracts quantifiable geometrical parameters (area, perimeter, shape factors) from defect images and uses these parameters in mathematical models to predict defect evolution, ensuring consistent results regardless of who performs the analysis.
Solution Approach 2:
The system replaces the human expert's interpretive process with automated image processing algorithms and mathematical models. The consistent application of these algorithms to extract geometrical parameters and predict defect evolution eliminates the variability inherent in human interpretation while maintaining reliable predictions.
3Measurement precision
If bearing pictures are sent to experts for interpretation, then defect analysis can be performed, but the bearing must be sent to expertise centers extending machine unavailability time
Solution Approach 1:
The patent implements an automated defect analysis system that can process bearing images on-site using a computer or mobile device. The system extracts geometrical parameters and predicts defect evolution locally, eliminating the need to transport bearings to expertise centers and thereby maintaining machine availability while ensuring accurate defect detection.
Solution Approach 2:
The system enables on-site self-diagnosis of bearing defects by processing images and predicting defect evolution using embedded algorithms and models. This self-service capability allows immediate defect analysis without external expert intervention, keeping the machine operational and reducing downtime.
4Measurement precision
If traditional neuronal networks are used to identify defects from pictures, then various surface features can be taken into account, but the network only identifies defects and does not predict their evolution
Solution Approach 1:
The patent combines defect identification and defect evolution prediction into a single integrated system. The image processing component identifies defects and extracts geometrical parameters, while the mathematical models use these parameters to predict defect evolution. This merging of functions provides both accurate identification and predictive capability, enhancing the system's versatility.
Solution Approach 2:
The geometrical parameters (area, perimeter, shape factors) extracted from defect images serve as intermediaries between the image identification stage and the defect evolution prediction stage. These parameters bridge the gap between visual defect detection and predictive analysis, enabling the system to both identify defects accurately and forecast their future development.
Data Source
AI summary
A method for predicting the evolution of a defect of a bearing includes identifying a defect of the bearing and extracting geometrical parameters of the identified defect by a trained deep learning algorithm from a picture of the bearing and further includes predicting an evolution of the identified defect of the bearing from a type of the identified defect and the extracted geometrical parameters of the identified defect, from operating parameters of the bearing and from a model of the bearing. Also a device for performing the method.