Bearing Intelligent Fault Diagnosis Based on Convolutional Neural Networks

The traditional intelligent identification method requires a complex feature extraction process and much diagnosis experience, considering the characteristics of one dimension of bearing vibration signals, a new method of intelligent fault diagnosis based on 1-dimensional convolutional neural network is presented. This method automatically extracts features from frequency domain signals and avoids artificial feature selection and feature extraction. The proposed method is validated on bearing benchmark datasets, these datasets are collected in different fault location, different health conditions and different operating conditions. The result shows that the proposed method can not only adaptively obtain representative fault features from the datasets, but also achieve higher diagnosis accuracy than the existing methods.

Energy Efficient Strategy Development of Steam Turbine through Vibration Reduction Using ANN and SVM Approaches

The energy efficiency of a power plant is largely determined by the vibrations of bearings that hold the shaft rotating at high speed which need to be critically controlled. This study presents the relative vibration modeling of a shaft bearing that is installed in a 660 MW supercritical steam turbine system. The operational data in raw form after being cleaned using machine learning based visualization and extensive data processing helped in training and validation of SVM and ANN models which are then compared by external validation tests. The model with best results is then used for the simulations of constructed operating scenarios. The ANN has been further tested for the complete operational load range (353 MW to 662 MW) which predicted the reduction in relative vibrations. Moreover, the validated ANN model has been used to develop many strategies of vibration reduction which helped in achieving more than 4% reduction in relative vibrations. Subsequently, an operational strategy that predicts a significant reduction in the bearing vibration levels is selected. For confirmation of the accuracy of prediction by ANN process model, the selected strategy has been used with the actual power plant. This assures the significant reduction of bearing vibration less than the alarm limit.

Booster Bearing Failure Reduction Through a Novel Thermodynamic Analysis

The gas booster station of a steel works has experienced excessive bearing failures since commissioning over two decades ago. This station was designed with redundancy, allowing for automatic switch-over between two gas booster fans. Bearing failures were observed, on average once every 15.7 days, with instances where both fans experienced simultaneous downtime. Booster failures resulted in regular station downtime, preventing Coke Oven Gas (COG) transport to an end user. This flammable by-product is used as a heat source and all unutilized volumes are flared, resulting in energy wastages. Furthermore, the absence of COG increases Natural Gas (NG) usage, procured at a cost. Traditional root cause analysis techniques failed to identify the cause of these excessive bearing failures. However, multiple in-depth data analysis studies resulted in a thermodynamic investigation, exposing liquid and solid particles within the COG to be responsible for the failures. This allowed for the design of an in-line particle collector, eliminating excessive failures. Following the particle collector installation, only two strategic bearing changes took place over the next 41 weeks, with reduced bearing vibration levels compared to before. The station experienced no failure downtime during this period, resulting in reduced COG flaring and thus improved energy utilization.

On the interpretation of deep learning models in bearing vibration diagnostics

In recent years, data-driven techniques such as deep learning (DL), have been widely represented in the literature in the field of bearing vibration condition monitoring. While these approaches achieve excellent performance in detecting bearing faults on controlled laboratory datasets, there is little information available on their applicability to more realistic working conditions. One challenge of these data-driven approaches is that they can learn non-classical features unrelated to the physical defect, making their generalizability debatable. To overcome the challenge of generalizability in DL models, we aim to first understand the underlying representation that the network uses to classify different bearing defects. Having an interpretable DL model may give us hints on how to increase its applicability by, e.g., data augmentation, changing input representations or adapting model architectures. To benefit from advances in interpretability in DL methods from computer vision, we first transform the vibration signal into an image. We evaluate a common input transformation, namely the spectrogram. Subsequently, the representations that the network has learnt are evaluated. We use the Grad-CAM algorithm together with signal modifications to evaluate which parts of the input signal contribute to class attribution. Our results show that the network learns signal features related to the transfer path, the physical properties of the test setup, rather than picking up classical features having a physical relation with the defect. Given that a transfer path is very machine specific, this could be an explanation for the lack of scalability of DL methods. To improve the generalizability of DL methods on bearing vibration analysis, the competing dominant machine specific features should be eliminated from the input representation. These results highlight the importance of combining domain expertise with data-driven approaches.

Collection and processing of bearing vibration data for their technical condition classification by machine learning methods

An experimental research facility has been developed to receive vibration signals from mechanisms with installed rolling bearings. A control block for all equipment has been created. For the repeatability of the experiment, an external microcontroller with a programmed proportional-integral-derivative regulator was used. Experiments were carried out to obtain initial data for different types of bearings. The processed data were grouped and made publicly available for further research. It is proposed to solve the problem of emergency stop of the generator, arising during operation due to bearings worn, by recognizing the pre-emergency conditions of rotary rig based on the use of advanced machine learning techniques: to highlight the signs of vibration and build clusters according to the degree of worn.

An Intelligent Bearing Fault Diagnosis Method Based on SF-SVM

Rolling bearing, as a key component of rotating machinery, its health status directly determines the stability and reliability of the whole machine. The research on its intelligent diagnosis method has important engineering value and academic significance. However, due to actual engineering conditions, the types of bearing failures and the amount of data are limited. Aiming at the difficulty of extracting and selecting bearing vibration features under limited sample constraints, this pa-per proposes an intelligent fault diagnosis method of SF-SVM. On the basis of the short-time Fourier change, the L2 regularized sparse filter is used to extract the unsupervised feature of the bearing vibration time-frequency map. After obtaining the typical features of the bearing, the support vector machine is used for diagnosis.

Analysis on vibration response of bearing coupling faults in rotor-bearing system

It is well known that the coupled fault diagnosis of rotating machinery is a challenging task, which is mainly due to the complexity of the vibration signals and the interaction of multiple fault components. In order to study the coupling vibration characteristics from unbalance and bearing fault in rotor-bearing system, A dynamic model of the coupled fault is performed to explore the phenomena of multi-fault in the rotor-bearing system in this paper. An experimental study, including seven working conditions, is also examined to further indicate the vibration response of coupling fault with unbalance and bearing fault. The influence of the rotor unbalance on the bearing vibration characteristics and the vibration rules of the coupling fault are proposed in this paper.

A Bearing Fault Diagnosis Method Based on PAVME and MEDE

When rolling bearings have a local fault, the real bearing vibration signal related to the local fault is characterized by the properties of nonlinear and nonstationary. To extract the useful fault features from the collected nonlinear and nonstationary bearing vibration signals and improve diagnostic accuracy, this paper proposes a new bearing fault diagnosis method based on parameter adaptive variational mode extraction (PAVME) and multiscale envelope dispersion entropy (MEDE). Firstly, a new method hailed as parameter adaptive variational mode extraction (PAVME) is presented to process the collected original bearing vibration signal and obtain the frequency components related to bearing faults, where its two important parameters (i.e., the penalty factor and mode center-frequency) are automatically determined by whale optimization algorithm. Subsequently, based on the processed bearing vibration signal, an effective complexity evaluation approach named multiscale envelope dispersion entropy (MEDE) is calculated for conducting bearing fault feature extraction. Finally, the extracted fault features are fed into the k-nearest neighbor (KNN) to automatically identify different health conditions of rolling bearing. Case studies and contrastive analysis are performed to validate the effectiveness and superiority of the proposed method. Experimental results show that the proposed method can not only effectively extract bearing fault features, but also obtain a high identification accuracy for bearing fault patterns under single or variable speed.

Big Data Quantitative Risk Analysis Method for Machine Health Indicator Prediction

Various data-driven methods have been applied to predict machine health indicators especially in the field of prognostics. Machine health indicators reveal the condition of equipment and/or its components including bearings by monitoring their operation data such as frequency vibration. To aid the prediction of the machine health indicators, this study applies the BDQRA method to monitor the health of bearings as a component of the machine. The BDQRA method involves applying data compression techniques like feature extraction to the bearing vibration data, to extract the most important features like time-domain, frequency domain, and time–frequency domain features. Due to the complexity of the feature extraction process, this study proposes fast Fourier transformation for the data compression. This is followed by obtaining a time series profile of the bearing vibration data to analyse the health status of component bearing. It the uses change-point analysis to predict the period at which the bearing health deterioration is imminent. Since the bearing health deterioration could be due to the independent operation of a component bearing or through communication between the component bearing and other components (or bearings) within the process machinery, the method also applies the principle of interaction effect to investigate the contributions from the other components of the machinery to the health deterioration of the component bearing detected. The accuracy of the prediction of the point of imminent health deterioration of the component bearing is investigated by comparing the outcome of the BDQRA method with the outcome of other methods published in literature which have been applied to the dataset used in this study. The findings reveal the BDQRA method have comparative advantages to the methods used in the related studies.

Feature Extraction based on Generalized Permutation Entropy for Condition Monitoring of Rotating Machinery

Defective rotating machinery usually exhibits complex dynamic behavior. Therefore, feature representation of machinery vibration signals is always critical for condition monitoring of rotating machinery. Permutation entropy (PeEn), an adaptive symbolic description, can measure complexities of signals. However, PeEn may lack the capability to fully describe dynamics of complex signals since compressing all the information into a single parameter. Afterwards, multiscale PeEn (MPeEn) is put forward for coping with nonstationarity, outliers and artifacts emerging in complex signals. In MPeEn, a set of parameters serve to describe dynamics of a complex signal in different time scales. Nonetheless, an average procedure in MPeEn may withhold local information of a complex signal. To overcome deficiencies of PeEn and MPeEn, this paper proposes generalized PeEn (GPeEn) by introducing different time lags and orders into PeEn. In GPeEn, a complex signal is compressed into one matrix rather than a single parameter. Moreover, minimal, maximal and average values of a matrix obtained by GPeEn serve to briefly describe conditions of rotating machinery. Next, a numerical experiment proves that GPeEn outperforms PeEn and MPeEn in characterizing conditions of a Lorenz model. Subsequently, the performance of GPeEn is benchmarked against that of PeEn and MPeEn by investigating gear and roll-bearing vibration signals containing different types and severity of faults. The results show that the proposed method has a clear advantage over PeEn and MPeEn in condition monitoring of rotating machinery.