|
1.INTRODUCTIONBall bearings in many applications, have become the most critical mechanical constituent. We all encounter these daily as motors, turbines, compressors, and vehicles. To ensure that those machines work correctly, ball bearings reduce the friction between the moving parts. The potential consequences of ball bearing failure can be severe, such that drastic downtime, expensive repair, and even safety hazards can be improved by earlier and more accurate detection of faults when needed for equipment to run with efficiency and high reliability1,2. These conventional fault-detection techniques mostly rely on physical inspection and fundamental signal analysis and are inadequate. They turn out to be sometimes too cumbersome, laborious, and not capable of fault detection at its nascent stage. Besides, most of these techniques often fail to exhibit the complete dynamics of bearing faults under changing operational conditions. Therefore, there exists a current demand for more sophisticated diagnosis techniques having correct, efficient, and least manual interventional strategies in the field of diagnosis of bearing faults. Condition in the bearings significantly reflects on the performance and reliability of rotating machinery. Detecting faults in such vital components is essential to avoid breakdowns and ensure smooth running3-5. The performance and reliability of rotating machinery heavily depend on the condition of its bearings. Combined vibration analysis with artificial intelligence scripting is one of the most potential methods of fault diagnosis because it gives more accuracy and efficiency and thus is considered to be effective6-8. Zhang et al. (2015)9 also recommended an intelligent way for roller bearing fault diagnosis. The method implemented was an incremental support vector machine with a multivariable ensemble. This approach significantly improved the accuracy of the process of fault classification using vibration signals. Real-time data incrementally emerges in online monitoring systems, and the ensemble-based method deals with such an incremental nature of real-time data. In the same context, ANNs and wavelet transform techniques in rolling element bearings’ fault diagnosis have been used by Gunerkar et al. Gunerkar et al. (2019)10 for fault conditions through vibration signals can be clearly identified under the proposed method; this is because it adequately follows a given procedure. The ability of ANNs to effectively handle non-linear and complex data associated with vibration is evident through the analysis, in which it forms a potent tool for fault Almounajjed et al. (2021)11 highlighted a review with the main points of some state-of-the-art machine learning algorithms used for fault diagnosis tasks related to the isolation and classification problems in rolling bearings and industries, with emphasis on the application of AI techniques in processing and analyzing vibration signals associated with faults detection and condition monitoring. Now, we address the adaptability and flexibility of machine learning algorithms across varied industrial applications. In their paper, Senthilnathan et al. 202412, reviewed developments in methods for diagnostic imperfections of the spheric roller bearing. The authors elaborate on several techniques for signal processing and artificial intelligence to filter noise and enhance an accurate fault diagnosis. This literature will help us understand the current trends and challenges in the field. Despite considerable advances, several fault diagnosis issues remain. One of the main issues is noisy data and changing operating conditions. Acknowledging these limitations, hybrid models and ensemble methods based on multiple machine-learning methods are studied here for better diagnostics13-15. This paper reports an overall outcome of the vibration data characterization and estimation of fault size. It emphasizes that more robust output can be obtained when we combine various techniques. The main objective is to develop an effective and accurate methodology for fault diagnosis in ball bearings with vibration data and statistical feature extraction. The objectives of the work specifically include the extraction of various remarkable statistical features from vibration signals to find unique patterns related to different fault conditions, training and testing of a Random Forest model for bearing condition classification, and its evaluation in standard metric model terms and open discussion concerning feature importance and their correlation concerning the fault diagnosis test and final validation through a large-scale experiment together with possible improvements. The results and discussion and the conclusion follow the methodology sections. References and appendices bringing out supplementary information join the main sections of this paper. 2.METHODOLOGY2.1Experimental workThe bearings have been tested using the FALEX multispecies test bench, available in the Tekniker lab16. In situ, the design and manufacturing of bearing faults took place in the DLR lab. The FALEX test bench has undergone essential modifications to integrate an in-depth monitoring infrastructure of forces, speed, temperature, vibration sensors, and a data acquisition system. The bearing test rig Figure 1 is fitted with three accelerometers model PCB 356A32 to measure triaxial vibrations along the x-, y-, and z-axes. Acceleration data was gathered at a 25.6 kHz sampling frequency. The bearing under test was installed in the vertical position in the test rig, and a vertical axial load was given to the inner race to make a simulation typical of the load and speed conditions for electro-mechanical aerospace systems. The setup of the experimental test bench FALEX enabled in-process monitoring of various parameters. It ensured exact control and accurate acquisition due to the created bearing faults having further very small vibration signals, high-frequency measuring accelerometers were used for the measurements. A dataset of 60 RPM spindle speed with a 5.0 kN axial load was taken for both inner and outer race faults. 2.2Bearing faults and specificationsThe two bearings categories considered in the experimental setup were healthy and faulty. The healthy bearings were simulated as control parameters for comparison. An intentionally created crack in the inner and outer races of the bearing was considered a fault. The fault created imitates a few common defects in the bearings, so it alters the vibration patterns significantly. The details of the ball bearing specifications used for the experiments are stated in Table 1. Table 1.Ball bearing specifications and fault characteristic frequencies for bearing QJ212TVP.
2.3Fault descriptionsIn this study, various fault cases were introduced to the ball bearings to simulate real-world defects and evaluate the effectiveness of the proposed fault diagnosis method. The faults were introduced on both the inner and outer races of the bearings. Each fault was carefully designed and manufactured to ensure consistency and reliability in the experimental results. The test conditions for all cases involved a spindle operating at 60 RPM and an axial load of 5.0 kN. Below are the detailed descriptions of the fault cases: 2.3.1Fault on inner race
Below is an illustration of the typical faults introduced in the inner and outer races of the ball bearings: These fault cases provide a comprehensive range of defect severities, allowing for a thorough evaluation of the fault diagnosis method’s ability to detect and classify various stages of bearing wear and damage. The consistent test conditions of 60 RPM spindle speed and 5.0 kN axial load ensure that the results are reliable and comparable across different fault scenarios. 2.4Data collection and feature extractionThe data collection process was meticulously planned and executed to ensure the reliability and accuracy of the vibration measurements. The experiments were conducted using the FALEX multi-specimen test bench17, equipped with a comprehensive monitoring infrastructure, including force, speed, temperature, and vibration sensors18. The bearing test rig was fitted with three accelerometers (model PCB 356A32)19 to measure triaxial vibrations along the x- y-, and z-axes at a sampling frequency of 25.6 kHz. For this study, only the x-axis measurements were used. Feature extraction is a critical step in the process of fault diagnosis as it involves transforming raw vibration data into a set of meaningful features that can be used to train machine learning models20-22. In this study, several statistical features were extracted from the timedomain vibration signals to capture the characteristics of the signals. The following statistical features were extracted from the vibration data16,23 (Table 2): Table 2.Statistical features extraction formula
where xi is a signal for i=1, 2, N, N is the number of data points. 2.4.Machine learning model and performanceThe experiments were performed using the Random Forest classifier for fault diagnosis, which has proven to acceptably manage the effects of noise and high dimensionality in data7,14. Random Forest is an ensemble learning technique that constructs a set of decision trees during training and then aggregates the votes from different trees to output the class chosen by the majority of the individual trees24. It has been selected for its effectiveness in avoiding overfitting maltreatment common to individual decision trees, working well with a large dataset, and handling a mix of numerical and categorical features. The extracted statistical features from the vibration data consisted of mean, median, standard deviation, minimum value, maximum value, range, skewness, kurtosis, and dominant frequency. The Random Forest model’s key hyperparameters, including the number of trees, maximum depth, and minimum samples per leaf, were optimized by grid search in which we trained the model on the available data and then a validation set was used to figure out the best combination of hyperparameters to be optimal for making predictions. The accuracy, precision, and recall with an F1-score realized to evaluate the Random Forest model were pointed out in the text, showing fully adequate performance for the classification of bearing conditions. 3.RESULTS AND DISCUSSION3.1Vibration signals for healthy and faulty statesIn a more visual and demonstrative manner, raw vibration signals were processed to highlight the differences between healthy and faulty bearings25,26. Figure 3 depicts the vibration signals of a healthy bearing and a defective bearing with a fault on the inner ring (Fault 1). The vibration signal from the healthy bearing is characterized by relatively low amplitude with stable patterns, indicative of smooth operation. In contrast, the amplitude of the faulty bearing is much higher with variable patterns, reflecting the imposed fault. These visual differences corroborate the statistical feature analysis and the model’s ability to distinguish between healthy and faulty states, aligning with previous studies on vibration signal analysis for fault detection. The vibration signal of the healthy bearing exhibits relatively low amplitude and consistent patterns, indicating smooth operation. In contrast, the faulty bearing’s signal shows higher amplitude and irregular patterns, reflecting the presence of the fault. These visual differences corroborate the statistical feature analysis and the model’s ability to distinguish between healthy and faulty states, supporting the findings of prior studies on vibration signal analysis for fault detection. 3.2Feature analysis using descriptive statisticsStatistical features extracted from vibration signals provide valuable information about the nature of data from both healthy and faulty bearings. Examining these features reveals various fault conditions with distinct patterns and variability. Descriptive statistics represent the data’s distribution and variability, crucial for effective fault diagnosis16. From Figure 4 can be observed the correlation matrix analysis was conducted to investigate the relationships among various features, aiding in selecting the most informative ones for the fault diagnosis model. Significant observations from the correlation matrix include:
These insights are crucial for selecting the most informative features for the fault diagnosis model, aligning with previous studies on feature selection in vibration analysis that underscore the importance of such analyses. 3.3Model performanceThe performance of the Random Forest model was evaluated using several metrics, including accuracy, precision, recall, and F1-score. The results are summarized in Table 3. Table 3.Model performance metrics.
The provided metrics show the classification of all the classes and different fault conditions by the model. This high level of accuracy, reinforced by appropriately balanced precision and recall scores, points to the expected excellent capacity of the model. It also fits very well with other studies conducted within a similar area, which uses machine learning in the same capacity for fault diagnosis. 4.CONFUSION MATRIX ANALYSISThe confusion matrix in Figure 5 elicits detailed information about the classification performance of the model by exposing both the correctly and mistakenly detected instances for every class. The power of the model is evaluated through the confusion matrix, which defines the weaknesses of the model. It is obvious that the Random Forest model performed well and, hence, is suitable for diagnosing ball bearings. From the analysis of the correlation matrix, it is easy to identify the most informative features; the mean, median, skewness, and kurtosis are those most contributing to the precision of the model. Through detailed analysis using the confusion matrix, the model’s performance in various fault conditions has been presented, and it also gives an idea about further refining areas of the model. This is congruent with previous research work, wherein it has been mentioned that feature selection and optimization of the model are very relevant to machine learning-based fault diagnosis28,29. 5.CONCLUSIONSThis was an excellent way of presenting material in fault diagnosis involving vibration data and statistical feature extraction for ball bearings. The outstanding findings and contributions of this research are as follows:
In this paper, the effectiveness of statistical features and Random Forest has been validated for diagnosing faults in ball bearings. The results obtained herein form a sound basis for further development and improvement of more advanced fault diagnosis to be real-time for enhancing the predictive maintenance and operation reliability of mechanical systems generally. ACKNOWLEDGMENTSThe authors wish to thank Mustansiriyah University and their afflation (www.uomustansiriyah.edu.iq) and Al Hikma University College, Baghdad, Iraq. REFERENCESMezni, Z., Delpha, C., Diallo, D. and Braham, A.,
“Performance of bearing ball defect classification based on the fusion of selected statistical features,”
Entropy, 24
(9), 1251
(2022). https://doi.org/10.3390/e24091251 Google Scholar
Altaf, M., Akram, T., Khan, M. A., Iqbal, M. and Hsu, C. H.,
“A new statistical features-based approach for bearing fault diagnosis using vibration signals,”
Sensors, 22
(5),
(2022). https://doi.org/10.3390/s22052012 Google Scholar
gnosis of industrial robot gears based on discrete wavelet transform and artificial neural network,”, Insight-Non-Destructive Testing and Condition Monitoring, 58
(4), 179
–186
(2016). https://doi.org/10.1784/insi.2016.58.4.179 Google Scholar
Jaber, A. A. and Bicker, R.,
“Development of a condition monitoring algorithm for industrial robots based on artificial intelligence and signal processing techniques,” International,”
Journal of Electrical & Computer Engineering, 81
(2), 996
–1009
(2018). Google Scholar
Dubaish, A. A. and Jaber, A. A.,
“Comparative Analysis of SVM and ANN for Machine Condition Monitoring and Fault Diagnosis in Gearboxes,”
Mathematical Modelling of Engineering Problems, 11
(4),
(2024). https://doi.org/10.18280/mmep Google Scholar
Ogaili, A. A. F., Hamzah, M. N. and Jaber, A. A.,
“Enhanced fault detection of wind turbine using extreme gradient boosting technique based on nonstationary vibration analysis,”
Journal of Failure Analysis and Prevention, 24 877
–895
(2024). https://doi.org/10.1007/s11668-024-01894-x Google Scholar
Ogaili, A. A. F., Jaber, A. A. and Hamzah, M. N.,
“Statistically optimal vibration feature selection for fault diagnosis in wind turbine blade,”
International Journal of Renewable Energy Research (IJRER), 13
(3), 1082
–1092
(2023). Google Scholar
Ogaili, A. A. F., Jaber, A. A. and Hamzah, M. N.,
“A methodological approach for detecting multiple faults in wind turbine blades based on vibration signals and machine learning,”
Curved and Layered Structures, 10
(1), 20220214
(2023). https://doi.org/10.1515/cls-2022-0214 Google Scholar
Zhang, X., Wang, B. and Chen, X.,
“Intelligent fault diagnosis of roller bearings with multivariable ensemble-based incremental support vector machine,” Knowl. Based,”
Syst., 89 56
–85
(2015). Google Scholar
Gunerkar, R. S., Jalan, A. K. and Belgamwar, S. U.,
“Fault diagnosis of rolling element bearing based on artificial neural network,”
Journal of Mechanical Science and Technology, 33
(2), 505
–511
(2019). https://doi.org/10.1007/s12206-019-0103-x Google Scholar
Almounajjed, A., Sahoo, A. K., Kumar, M. K. and Alsebai, M. D.,
“Investigation techniques for rolling bearing fault diagnosis using machine learning algorithms,”
in 2021 5th International Conference on Intelligent Computing and Control Systems (ICICCS),
1290
–1294
(2021). Google Scholar
Senthilnathan, N., Babu, T. N., Varma, K. S. D., Rushmith, S., Reddy, J. A., Kavitha, K. V. N. and Prabha, D. R.,
“Recent advancements in fault diagnosis of spherical roller bearing: a short review,”
Journal of Vibration Engineering & Technologies, 12
(4), 6963
–6977
(2024). https://doi.org/10.1007/s42417-024-01293-4 Google Scholar
Sánchez, R.-V., Lucero, P., Vásquez, R. E., Cerrada, M., Macancela, J.-C. and Cabrera, D.,
“Feature ranking for multi-fault diagnosis of rotating machinery by using random forest and KNN,”
Journal of Intelligent & Fuzzy Systems, 34 3463
–3473
(2018). https://doi.org/10.3233/JIFS-169526 Google Scholar
Vakharia, V., Gupta, V. K. and Kankar, P. K.,
“Efficient fault diagnosis of ball bearing using Relief and Random Forest classifier,”
Journal of the Brazilian Society of Mechanical Sciences and Engineering, 39 2969
–2982
(2017). https://doi.org/10.1007/s40430-017-0717-9 Google Scholar
Mohammed, K. A., Al-Sabbagh, M. N. M., Ogaili, A. A. F. and Al-Ameen, E. S.,
“Experimental analysis of hot machining parameters in surface finishing of crankshaft,”
Procedia Manufacturing, 20 265
–270
(2018). Google Scholar
Ismail, M., Windelberg, J., Bierig, A., Bravo, I., Brief, A. A.-D.,
“Ball bearing vibration data for detecting and quantifying spall faults,”
Data in Brief, 47 109019
(2023). https://doi.org/10.1016/j.dib.2023.109019 Google Scholar
Ismail, M. A. A., Bierig, A. and Sawalhi, N.,
“Automated vibration-based fault size estimation for ball bearings using Savitzky-Golay differentiators,”
24
(18), 4297
–4315
(2017). Google Scholar
Mohammed, K. A., Al-Sabbagh, M. N. M., Ogaili, A. A. F. and Al-Ameen, E. S.,
“Experimental analysis of hot machining parameters in surface finishing of crankshaft,”
Journal of Mechanical Engineering Research and Developments, 43
(4), 105
–114 Google Scholar
Ogaili, A. A. F., Jaber, A. A. and Hamzah, M. N.,
“Wind turbine blades fault diagnosis based on vibration dataset analysis,”
Data Brief, 49 109414
(2023). https://doi.org/10.1016/j.dib.2023.109414 Google Scholar
Al-Haddad, L. A. and Jaber, A. A.,
“Improved UAV blade unbalance prediction based on machine learning and ReliefF supreme feature ranking method,”
Journal of the Brazilian Society of Mechanical Sciences and Engineering, 45
(9), 463
(2023). https://doi.org/10.1007/s40430-023-04386-5 Google Scholar
Al-Haddad, L. A. and Jaber, A. A.,
“An intelligent fault diagnosis approach for multirotor UAVs based on deep neural network of multi-resolution transform features,”
Drones, 7
(2), 82
(2023). https://doi.org/10.3390/drones7020082 Google Scholar
Al-Haddad, L. A., Jaber, A., Neranon, P. and Al-Haddad, S. A.,
“Investigation of frequency-domain-based vibration signal analysis for UAV unbalance fault classification,”
Engineering and Technology Journal, 41
(7), 1–9
(2023). https://doi.org/10.30684/etj.2023.137412.1348 Google Scholar
“QJ212TVP FAG bearings,
(2024) https://www.fagbearing.cc/FAG-bearings/QJ212TVP_FAG_50194.html > (8 June ). 2024). Google Scholar
Breiman, L.,
“Random forests,”
Mach Learn, 45
(1), 5
–32
(2001). https://doi.org/10.1023/A:1010933404324 Google Scholar
Ogaili, A. A., Hamzah, M. N., Jaber, A. Ab. and Ghane, E.,
“Application of discrete wavelet transform for condition monitoring and fault detection in wind turbine blades: an experimental study,”
Engineering and Technology Journal, 42
(1), 104
–116
(2024). Google Scholar
Ogaili, A. A. F., Jaber, A. A. and Hamzah, M. N.,
“Wind turbine blades fault diagnosis based on vibration dataset analysis,”
Data Brief, 49 109414
(2023). https://doi.org/10.1016/j.dib.2023.109414 Google Scholar
A. A. Shandookh, A. A. Farhan Ogaili, and L. A. Al-Haddad,
“Failure analysis in predictive maintenance: Belt drive diagnostics with expert systems and Taguchi method for unconventional vibration features,”
Heliyon, 10
(13), e34202
(2024). https://doi.org/10.1016/j.heliyon.2024.e34202 Google Scholar
Ogaili, A. A. F., Mohammed, K. A., Jaber, A. A., & Al, A. E. S.,
“Automated wind turbines gearbox condition monitoring: A comparative study of machine learning techniques based on vibration analysis,”
FME Transactions, 52(3), 471
–485
(2024). https://doi.org/10.5937/fme2403471O Google Scholar
Al-Haddad, L. A., Al-Muslim, Y. M., Hammood, A. S., Al-Zubaidi, A. A., Khalil, A. M., Ibraheem, Y., Imran, H. J., Fattah, M. Y., Alawami, M. F. and Abdul-Ghani, A. M.,
“Enhancing building sustainability through aerodynamic shading devices: an integrated design methodology using finite element analysis and optimized neural networks,”
Asian Journal of Civil Engineering, 25
(5), 4281
–4294
(2024). https://doi.org/10.1007/s42107-024-01047-3 Google Scholar
|