Rolling bearings are widely used in industrial machines, and their deterioration can lead to the failure of rotating machinery. To diagnose the condition of rolling bearings, AI-based methods are commonly employed. Among these methods, hybrid neural network (NN)-based approaches have been found to achieve the best diagnosis results. Typically, raw data is collected from accelerometers mounted on the machine housing. However, the effectiveness of each signal in diagnosing bearing defects depends heavily on the location of the corresponding accelerometer. In this study, a novel hybrid CNN-MLP model-based diagnostic method is proposed for rolling bearing diagnostics. This method combines mixed input data to detect and localize bearing defects using acceleration data from a shaft-mounted wireless acceleration sensor. The researchers compared the performance of the hybrid model with that of CNN and MLP models operating separately. The experimental results demonstrate that the hybrid CNN-MLP model outperforms both CNN and MLP models individually. It achieves a high detection accuracy rate of 99.6% for bearing faults, compared to 98% for CNN models and 81% for MLP models. In terms of methodology, various analysis techniques are commonly used in fault recognition for rolling bearings. Time-domain analysis calculates statistical parameters such as RMS and kurtosis, while frequency-domain analysis isolates key frequency components using tools like fast Fourier transform (FFT). Time-frequency analysis extends frequency-domain analysis to non-stationary vibration signals through methods like short-term Fourier transform, wavelet transform, empirical mode decomposition (EMD), or Hilbert-Huang transform (HHT). For fault recognition purposes, extracted features from these analyses serve as inputs to machine learning techniques. Different machine learning tools have been utilized in previous studies, including k-nearest neighbors (k-NN) method, naive Bayes algorithm, Bayes network algorithm, support vector machine approach, random forests algorithm, and artificial neural networks. Deep neural networks have shown great potential in mechanical fault diagnosis, with convolutional neural networks (CNN) being particularly effective in bearing fault diagnosis. Researchers have successfully trained one-dimensional CNN models using raw vibration signals.
- Rolling bearings are widely used in industrial machines and their deterioration can lead to machinery failure.
- AI-based methods, specifically hybrid neural network (NN)-based approaches, are commonly used to diagnose the condition of rolling bearings.
- Raw data is collected from accelerometers mounted on the machine housing for diagnosis.
- The effectiveness of each signal in diagnosing bearing defects depends on the location of the accelerometer.
- A novel hybrid CNN-MLP model-based diagnostic method is proposed for rolling bearing diagnostics, which combines mixed input data and wireless acceleration sensor data.
- The hybrid CNN-MLP model outperforms both CNN and MLP models individually, achieving a high detection accuracy rate of 99.6% for bearing faults.
- Various analysis techniques are commonly used in fault recognition for rolling bearings, including time-domain analysis, frequency-domain analysis, and time-frequency analysis.
- Extracted features from these analyses serve as inputs to machine learning techniques for fault recognition purposes.
- Different machine learning tools have been utilized in previous studies, including k-nearest neighbors (k-NN) method, naive Bayes algorithm, Bayes network algorithm, support vector machine approach, random forests algorithm, and artificial neural networks.
- Convolutional neural networks (CNN) have shown great potential in mechanical fault diagnosis, particularly in bearing fault diagnosis.
Rolling bearings are used in machines and can break if they get damaged. AI-based methods, which are computer programs that think like humans, are used to check if rolling bearings are working properly. Data is collected from sensors on the machine to diagnose any problems with the bearings. The location of the sensor affects how well it can detect problems with the bearings. A new method using a mix of data and wireless sensors has been developed to diagnose bearing problems. This method is better than other methods, with a 99.6% accuracy rate for detecting bearing faults. Different techniques like analyzing time and frequency are used to recognize faults in rolling bearings. Information from these analyses is used by computers to recognize faults in the bearings. Different computer tools have been used in previous studies to recognize faults, like k-nearest neighbors, naive Bayes algorithm, support vector machine approach, random forests algorithm, and artificial neural networks. Convolutional neural networks have shown promise in diagnosing mechanical faults, especially in bearing problems."
Definitions
- Rolling Bearings: These are parts that help machines move smoothly and can break if they get damaged.
- AI-based methods: These are computer programs that use artificial intelligence to think like humans and solve problems.
- Diagnose: This means figuring out what is wrong with something or someone.
- Accelerometers: These are sensors that measure how fast something is moving or shaking.
- Defects: These are things that are wrong or broken in something.
- Hybrid
Rolling Bearing Diagnostics with Hybrid CNN-MLP Model
Rolling bearings are a critical component of industrial machines, and their deterioration can lead to the failure of rotating machinery. To diagnose the condition of rolling bearings, AI-based methods are commonly employed. Among these methods, hybrid neural network (NN)-based approaches have been found to achieve the best diagnosis results. In this article, we discuss a novel hybrid CNN-MLP model-based diagnostic method proposed for rolling bearing diagnostics.
Data Collection
Raw data is typically collected from accelerometers mounted on the machine housing. However, the effectiveness of each signal in diagnosing bearing defects depends heavily on the location of the corresponding accelerometer. In this study, acceleration data from a shaft-mounted wireless acceleration sensor was used as input for detecting and localizing bearing defects.
Hybrid Model Architecture
The proposed hybrid CNN-MLP model combines mixed input data to detect and localize bearing defects using acceleration data from a shaft-mounted wireless acceleration sensor. The researchers compared the performance of the hybrid model with that of CNN and MLP models operating separately. The experimental results demonstrate that the hybrid CNN-MLP model outperforms both CNN and MLP models individually, achieving a high detection accuracy rate of 99.6% for bearing faults compared to 98% for CNN models and 81% for MLP models.
Analysis Techniques Used in Fault Recognition
Various analysis techniques are commonly used in fault recognition for rolling bearings: time domain analysis calculates statistical parameters such as RMS and kurtosis; frequency domain analysis isolates key frequency components using tools like fast Fourier transform (FFT); time frequency analysis extends frequency domain analysis to non stationary vibration signals through methods like short term Fourier transform, wavelet transform, empirical mode decomposition (EMD), or Hilbert Huang transform (HHT). For fault recognition purposes extracted features from these analyses serve as inputs to machine learning techniques such as k nearest neighbors (k NN) method, naive Bayes algorithm, Bayes network algorithm support vector machine approach random forests algorithm artificial neural networks etc., Deep neural networks have shown great potential in mechanical fault diagnosis with convolutional neural networks being particularly effective in bearing fault diagnosis Researchers have successfully trained one dimensional CNN models using raw vibration signals
Conclusion
In conclusion this paper has discussed a novel hybrid Convolutional Neural Network - Multi Layer Perceptron based diagnostic method which has been proven to be more accurate than either individual technique when it comes to diagnosing rolling bearing faults It achieves an impressive detection accuracy rate of 99 6%, compared to 98% for Convolutional Neural Networks alone 81% for Multi Layer Perceptrons alone This research provides valuable insight into how deep learning algorithms can be effectively applied towards improving mechanical system reliability
