Neural Network Fault Diagnosis Research Articles

A comparative study on long short-term memory and gated recurrent unit neural networks in fault diagnosis for chemical processes using visualization

Recurrent neural networks (RNNs), particularly those with gated units, such as long short-term memory (LSTM) and gated recurrent unit (GRU), have demonstrated clear superiority in sequence modeling. In this study, the fault diagnosis performance and classification mechanisms of basic LSTM and GRU were investigated to provide comparative information about suitable fault diagnosis models in chemical processes. Visualization techniques were used to interpret the behavior of LSTM and GRU when performing fault diagnosis in the Tennessee Eastman process (TEP). The analysis indicated that input and output gates were the two main gates of the LSTM model to filter information, while the forget gate of LSTM might be insignificant because it was usually opened to allow information to pass through. Therefore, the GRU model separated the faults better, especially Fault 15, and it provided more promising fault diagnosis performance compared to the LSTM model. The diagnosis accuracy for Fault 15 increased from 63%, while using the LSTM model, to 76% while using the GRU model. The simulation results of the TEP indicated that the GRU neural network in this study was superior to the LSTM neural network.

Unilateral alignment transfer neural network for fault diagnosis of aircraft engine

Data-driven fault diagnosis for aircraft engine has recently been widely studied, most of which utilized traditional machine learning tools. However, these approaches often assumed that the data used for training and testing come from the identical distribution, which is unpractical in practice, because even for the same type of engine, there may be some discrepancies in the data collected between different engine individuals. This idealized assumption may confine these promising data-driven technologies to well-design experimental environments instead of applying them to practical applications. Meanwhile, the research on data-driven engine fault diagnosis methods based on transfer learning is relatively limited. To overcome this issue, this paper proposes a methodology using unilateral alignment strategy to address the issue of cross domain engine gas path fault diagnosis, where the training data extracted from source domain and the testing data extracted from target domain are assumed to come from different distributions. More importantly, the proposed methodology is capable of maintaining the inter-class relationship of source domain, instead of aligning the discrepancies for source and target domain by mapping both of their feature spaces into an unknown intermediate space, because the label space between two domains may be different, and the label set of target domain might be only a subset of that of source domain. Meanwhile, both the marginal and conditional distribution discrepancies are considered. Finally, the proposed approaches are evaluated by extensive experiments on engine cross domain fault diagnosis, including hybrid transfer experiments, complete transfer experiments and missing class experiments, and the overall results demonstrate the feasibility of the proposed methodology.

Research on the application of artificial intelligence method in automobile engine fault diagnosis

The application of artificial intelligence methods in fault diagnosis is becoming more and more extensive, and exploring and researching intelligent fault diagnosis methods for automobile engines is also a hot spot in the field of automotive engineering. Different machine learning methods have their own advantages and disadvantages. By extracting different characteristic parameters and optimizing the combination of multiple algorithms, faster and stable diagnosis results can be achieved, so that faults can be discovered and repaired in time. Aiming at the potential fluctuation and impact characteristics of vibration plus signal caused by different failure states of automobile engines, this paper proposes two engine fault identification methods using vibration acceleration signals as diagnostic parameters. They are Cross Validation -Support Vector Machine (CV-SVM)and Particle Swarm Optimization-Probabilistic Neural Network (PSO-PNN) engine fault identification methods. The advantages and disadvantages of the two methods are compared and analyzed. Obtain the amplitude corresponding to the frequency multiplication of the vibration acceleration signal through the spectrum analysis method, which is used as the main component of the input feature vector, and establish the SVM fault diagnosis model combined with the cross-validation method (CV); In addition, multiple one-dimensional arrays composed of time-domain signals are directly used as input feature vectors, and the particle swarm optimization (PSO) parameter optimization is used to obtain the best Probabilistic Neural Network(PNN) fault diagnosis model. The results show that both the CV-SVM (small sample) method and the PSO-PNN method (large sample) can realize the identification and diagnosis of the established engine fault type. The CV-SVM method has more advantages in diagnostic fault tolerance, but the PSO-PNN method can simplify the process of feature sample preparation, save a lot of manual feature extraction tasks, and is more convenient in practical application.

A Novel Transformers Fault Diagnosis Method Based on Probabilistic Neural Network and Bio-Inspired Optimizer.

Since it is difficult for the traditional fault diagnosis method based on dissolved gas analysis (DGA) to meet today’s engineering needs in terms of diagnostic accuracy and stability, this paper proposes an artificial intelligence fault diagnosis method based on a probabilistic neural network (PNN) and bio-inspired optimizer. The PNN is used as the basic classifier of the fault diagnosis model, and the bio-inspired optimizer, improved salp swarm algorithm (ISSA), is used to optimize the hidden layer smoothing factor of PNN, which stably improves the classification performance of PNN. Compared with the traditional SSA, the sine cosine algorithm (SCA) and disruption operator are introduced in ISSA, which effectively improves the exploration capability and convergence speed. To verify the engineering applicability of the proposed method, the ISSA-PNN model was developed and tested using sensor data provided by Jiangxi Province Power Supply Company. In addition, the method is compared with machine learning methods such as support vector machine (SVM), back propagation neural network (BPNN), multi-layer perceptron (MLP), and traditional fault diagnosis methods such as the international electrotechnical commission (IEC) ratio method. The results show that the proposed method has a strong learning ability for complex fault data and has advantages in accuracy and robustness compared to other methods.

Novel topology convolutional neural network fault diagnosis for aircraft actuators and their sensors

The variable working conditions and frequent turns make the aircraft actuator system prone to failure, seriously threatening flight safety. The identification of the airplane actuator system is critical for flight decisions and safety. Most fault diagnosis methods of actuators only focus on the actuators themselves, ignoring the disturbance caused by the fault of the actuator position sensor, which may easily lead to wrong decisions. In order to distinguish the actuator fault from its position sensor fault and identify the fault type accurately, an offline diagnosis method of convolutional neural network (CNN) with novel topology for processing time series is proposed. A new shift layer is added after the input layer, which avoids the loss of a large number of features due to the direct connection between the time series and the convolutional layer. A local topological network learning complex pattern with inception module is designed to improve the diagnostic accuracy in different working conditions. The wide residual structure is introduced to expand the convolutional channel, which allows the network features at the bottom level to propagate directly to the top level to prevent network degradation. Simulation results show that this method can accurately diagnose the actuator fault and its position sensor, with an average accuracy of 96.8%. Compared with the current mainstream data-driven methods, the precision and recall are increased by 6.3% and 6.7% respectively.

A Multi-Input and Multi-Task Convolutional Neural Network for Fault Diagnosis Based on Bearing Vibration Signal

Bearing fault diagnosis is essential for the safe and stable operation of rotating machinery. Existing methods use signals from a single dimension, limiting diagnostic generality and accuracy. To address these limitations and make improved use of signal features from multiple dimensions, a novel convolutional neural network model with multi-dimensional signal inputs and multi-dimensional task outputs called MIMTNet is proposed. First, frequency domain signals and a time frequency graph are obtained by using the short-time Fourier transform and a wavelet transform to process original time domain signals simultaneously. Then, the time domain signals, the frequency domain signals, and the time frequency graph are fed into the model and a special aggregation is performed after extracting features from the three corresponding branches. Finally, the outputs of the three-dimensional tasks are acquired by different full connection layers to process the aggregated features of bearing position, damage location within the bearing, and the damage size. Two common bearing vibration signal datasets are used to verify the generalization ability of our proposed method. And experimental results show that the proposed method effectively improves the bearing diagnosis capability of the deep learning model.

An Integrated Approach Based on Improved CEEMDAN and LSTM Deep Learning Neural Network for Fault Diagnosis of Reciprocating Pump

The reciprocating pump plays an important role in the petrochemical industry procedure, it is crucial in ensuring the systematic safety and stability. Since the useful feature information of the vibration signal from the reciprocating pump tends to be overwhelmed by the background ingredients, it is tough to realize the recognition on typical modes. Aiming at the extraction of reciprocating mechanical fault features and mode recognition, this paper proposes Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise and LSTM (Long Short-Term Memory) deep neural network algorithm. Firstly, the IMF components are obtained by decomposing the vibration signals from the reciprocating pump with the Improved CEEMDAN algorithm, in which the key parameter β <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">k</sub> is improved and redefined, for optimizing SNRs (Signal Noise Ratio) of the IMF (Intrinsic Mode Function) components. Then the corresponding singular spectral entropy is calculated and the feature vector is constructed. The classification modal based on LSTM deep network is developed in the data dividing-training and the final mode recognition process. The study shows that the proposed method can effectively extract the fault features of vibration signal of the reciprocating pump, and the testing modes could be accurately recognized with the developed classification model.

The Fault Diagnosis of Diesel Fuel Supply System Based on BP Neural Network Optimized by Genetic Algorithm

As an important power source of diesel engine, the fuel supply system has the characteristics of high failure rate and being difficult to be detected, which is critical to detect and diagnose the fault for the reliable and stable operation of diesel engine. In the paper, principal component analysis (PCA) was used to process the initial data, back-propagation (BP) neural network was used to diagnose fault of diesel engine, BP diagnosis efficacy was analyzed, then the BP neural network fault diagnosis system was optimized by being introduced in the genetic algorithm , and the comparison results of training error and diagnostic analysis showed that: the fault diagnosis of BP neural network optimized by genetic algorithm is more accurate than that of traditional BP, which can effectively improve the fault diagnosis accuracy and reduce the training time.

Research on Rotating Machinery Fault Diagnosis Method Based on Energy Spectrum Matrix and Adaptive Convolutional Neural Network

Traditional intelligent fault diagnosis methods focus on distinguishing different fault modes, but ignore the deterioration of fault severity. This paper proposes a new two-stage hierarchical convolutional neural network for fault diagnosis of rotating machinery bearings. The failure mode and failure severity are modeled as a hierarchical structure. First, the original vibration signal is transformed into an energy spectrum matrix containing fault-related information through wavelet packet decomposition. Secondly, in the model training method, an adaptive learning rate dynamic adjustment strategy is further proposed, which adaptively extracts robust features from the spectrum matrix for fault mode and severity diagnosis. To verify the effectiveness of the method, the bearing fault data was collected using a rotating machine test bench. On this basis, the diagnostic accuracy, convergence performance and robustness of the model under different signal-to-noise ratios and variable load environments are evaluated, and the feature learning ability of the method is verified by visual analysis. Experimental results show that this method has achieved satisfactory results in both fault pattern recognition and fault severity evaluation, and is superior to other machine learning and deep learning methods.

Research on Fault Diagnosis Model of Transmission Line under Lightning Stroke Based on Neural Network

Transmission lines are an important part of the power system. In recent years, there have been more and more lightning trips, power outages and outages, which have become one of the hazards that threaten the stable operation of the power grid, and seriously endanger the reliability of the power grid operation. It is necessary to diagnose problems with transmission line faults. In this paper, after the information generated by the fault is processed by wavelet, it is input into the neural network fault diagnosis model as the input part, and the fault type of the transmission line is predicted and identified by the neural network operation, and the experiment is carried out through MATLAB simulation software. The prediction results show that the model is more accurate in fault diagnosis of transmission lines.

PNN and BP Neural Network Fault Diagnosis Research on Electronic Accelerator Pedal Detection

Background: The working state of electronic accelerator pedal directly affects the safety of vehicles and drivers. Effective fault detection and judgment for the working state of the accelerator pedal can prevent accidents. Methods: Aiming at different working conditions of electronic accelerator pedal, this paper used PNN and BP diagnosis model to detect the state of electronic accelerator pedal according to the principle and characteristics of PNN and BP neural network. The fault diagnosis test experiment of electronic accelerator pedal was carried out to get the data acquisition. Results: After the patents for electronic accelerator pedals are queried and used, the first measured voltage, the upper limit of first voltage, the first voltage lower limit, the second measured voltage, the upper limit of second voltage and the second voltage lower limit are tested to build up the data samples. Then the PNN and BP fault diagnosis models of electronic accelerator pedal are established. Six fault samples are defined through the design of electronic accelerator pedal fault classifier and the fault diagnosis processes are executed to test. Conclusion: The fault diagnosis results were analyzed and the comparisons between the PNN and the BP research results show that BP neural network is an effective method for fault detection of electronic throttle pedal, which is obviously superior to PNN neural network based on the experiment data.

Interpreting network knowledge with attention mechanism for bearing fault diagnosis

Condition monitoring and fault diagnosis of bearings play important roles in production safety and limiting the cost of maintenance on a reasonable level. Nowadays, artificial intelligence and machine learning make fault diagnosis gradually become intelligent, and data-driven intelligent algorithms are receiving more and more attention. However, many methods use the existing deep learning models directly for the analysis of mechanical vibration signals, which is still lack of interpretability to researchers. In this paper, a method based on multilayer bidirectional gated recurrent units with attention mechanism is proposed to access the interpretability of neural networks in fault diagnosis, which combines the convolution neural network, gated recurrent unit, and the attention mechanism. Based on the attention mechanism, the attention distribution of input segments is visualized and thus the interpretability of neural networks can be further presented. Experimental validations and comparisons are conducted on bearings. The results present that the proposed model is effective for localizing the discriminative information from the input data, which provides a tool for better understanding the feature extraction process in neural networks, especially for mechanical vibration signals.

Mathematical model and diagnosis of rotor fault in synchronous condenser

With the application of new energy in grid-connected system and the development of UHV DC transmission, the grid's requirements for reactive power regulation have gradually increased. Considering that, large-scale synchronous condensers have been put into use again. However, it is difficult to extract the characteristic signal of the inter-turn short-circuit fault in rotor windings of synchronous motors. In order to improve the stability of condensers, a certain relationship between the excitation current and the number of turns is derived using the Parker equation in the dq0 coordinate system, and the differential equation simulates the excitation current. Then the characteristic energy value of the fault signal is extracted through wavelet packet decomposition and reconstruction, and it is input to the RBF neural network for fault diagnosis. It is proved by MATLAB simulation that the diagnostic method proposed in this paper can effectively detect the degree of short-circuit faults between the turns of the rotor in condensers.

A novel normalized recurrent neural network for fault diagnosis with noisy labels

The early fault diagnosis is a kind of important technology to ensure the normal and reliable operation of wind turbines. However, due to the potential presence of noisy labels in health condition dataset and the weakly explanation of the deep neural network decisions, the performance of fault diagnosis is severely limited. In this paper, a framework called normalized recurrent neural network (NRNN) is proposed for noisy label fault diagnosis, in which the normalized long short-term memory is used to improve the training process and the forward crossentropy loss is introduced to handle the negative effect of noisy labels. The effectiveness and superiority of the proposed framework are verified by four datasets with different noisy label proportions. Meanwhile, the layer-wise relevance propagation algorithm is applied to explore the decision of framework and by visualizing the relevances of input samples to framework decisions, the NRNN does not treat samples equally and prefers signal peaks for classification decisions.

Fault Diagnosis of Gearbox in Multiple Conditions Based on Fine-Grained Classification CNN Algorithm

The use of the convolutional neural network for fault diagnosis has been a common method of research in recent years. Since this method can automatically extract fault features, it has played a good role in some research studies. However, this method has a clear drawback that the signals will be significantly affected by working conditions and sample size, and it is difficult to improve diagnostic accuracy by directly learning faults, regardless of working conditions. It is therefore a research orientation worthy of a diagnosis of high precision defect in various working conditions. In this article, using a fine-grained classification algorithm, the operating conditions of the object system are considered an approximate classification. A specific failure in different working conditions is considered a beautiful classification. Samples of different faults in different working conditions are learned uniformly and the common characteristics are extracted from the convolutional network so that different faults of different working conditions can simultaneously be identified on the basis of the entire sample. Experimental results show that the method effectively uses the set of samples of the working conditions of the variables to obtain the dual recognition of defects and specific working conditions and the accuracy of the recognition is significantly higher than the method of learning regardless of working conditions.

An adversarial denoising convolutional neural network for fault diagnosis of rotating machinery under noisy environment and limited sample size case

The rapid development of deep learning raises a new research area for condition monitoring and fault diagnosis of mechanical equipment recently. However, the amount of labeled fault samples is limited in industrial field, also the samples are filled with complex environmental noise. Thus, a model with strong generalization and robustness is required. To tackle these challenges, an adversarial denoising convolutional neural network (ADCNN) is proposed in this paper. Moving maximum is firstly applied to the frequency spectrum of the vibration signal to enhance the anti-noise performance of the training samples. Then the enhanced training samples are erased with dynamic probability to simulate noise interference. Meanwhile, adversarial training is utilized to expand the labeled samples until Nash equilibrium is reached. These processes improve the robustness and generalization of ADCNN, and avoid over-fitting with limited amount of labeled samples. In our two experiments, when signal to noise ratio(SNR) is -12dB, the ADCNN achieves a diagnosis accuracy of 94.05% and 94.47%, respectively. Besides, ADCNN can maintain a high diagnosis accuracy under limited sample size case, and has the best adaptation performance across different load domains. The comparison studies with respect to other models demonstrate the applicability and superiority of ADCNN.

Combined Optimization of Neural Network Fault Diagnosis Methods for Analog Circuits on Ships

Guo, S.; Li, H.; Wu, B.; Zhou, J.; Su, C., and Guan, C., 2020. Combined optimization of neural network fault diagnosis methods for analog circuits on ships. In: Zheng, C.W.; Wang, Q.; Zhan, C., and Yang, S.B. (eds.), Air-Sea Interaction and Coastal Environments of the Maritime and Polar Silk Roads. Journal of Coastal Research, Special Issue No. 99, pp. 158–164. Coconut Creek (Florida), ISSN 0749-0208.With increasing automation, electronic systems are becoming more widely used on ships. When electronic systems fail, it will greatly affect the safety of ship operation or even cause catastrophic accidents. Therefore, it is necessary to diagnose a circuit fault in order to repair the equipment in time. In electronic systems, more than 80% of the faults come from analog circuits. To obtain better fault diagnosis accuracy for analog circuits, neural network classification algorithms are often combined with optimization steps, such as circuit feature collection, feature engineering, regularization and so on, significantly affecting the accuracy of the classification algorithm, such as circuit feature collection, feature engineering, and regularization. However, the accuracy of analog circuit fault diagnosis is often not high in practical applications because of the tolerance of analog circuit components and other factors. To solve the above problems, this paper proposes a neural network combination optimization method based on hypothesis testing and applies it this method to the fault diagnosis of analog circuits on ships. This method uses hypothesis testing to combine and sort the optimization methods of each step in a neural network classification algorithm to obtain a model ranking based on the accuracy rate with mathematical statistics rules, thereby obtaining the optimal combination method. Finally, the validity of the method is verified by experiments.

An integrated method based on hybrid grey wolf optimizer improved variational mode decomposition and deep neural network for fault diagnosis of rolling bearing

• A parametric adaptive VMD method based on Hybrid Grey Wolf Optimizer is proposed. • The improved VMD method can effectively and significantly extract the subtle early fault feature. • SVD is a good tool for extracting the robust features. • DBN has higher correctness and stability for recognizing faults than other traditional classifiers. • This new fault diagnosis model (IVMD-SVD-DBN) has higher recognition accuracy than other combination methods. The selection of penalty parameters along with the number of components in the Variational Modal Decomposition (VMD) determines the decomposition effect to a large degree. In order to achieve the optimal selection of relevant parameters in VMD, an improved parametric adaptive VMD method based on Hybrid Grey Wolf Optimizer (HGWO) is proposed. First of all, taking the minimum local envelope entropy of modal components in the VMD as the optimization goal, HGWO algorithm is used to search for the optimal parameter combination in VMD. Then, the improved VMD is used to decompose the vibration signal to obtain modal components. For improving the stability of fault feature, the initial eigenmatrix composed of the key modal components are decomposed by Singular Value Decomposition (SVD), and the singular value is taken as the final eigenmatrix. Finally, the feature matrix is input to the Deep Belief Network (DBN) for learning and training, so as to realize the early fault diagnosis of rolling bearing. The comparative experimental analysis shows that the improved VMD method after parameter optimization can extract the early failure characteristics of rolling bearing more distinctly, and the fault diagnosis model based on this method has higher accuracy and application value.

Fault Diagnosis in Multi Phase Induction Machine using Mind Evolution Computation Algorithm Optimized Neural Network

This article proposes a new solution method for diagnosing faults in a multi phase induction motor using least mean square filter (LMS) and a new hybrid neural network with mind evolution computation algorithm. The entire procedure for teaching an artificial neural network (ANN) is popularly thought of among the toughest activities in system learning and also it has lately attracted lots of research workers. The proposed hybrid fault diagnosing method includes an efficient feature extractor based on LMS and a fault classifier based on a hybrid neural network. First, the LMS method is used to extract the effective features. The mind evolution computation algorithm is employed to train the neural network. The performance and efficiency of the presented hybrid neural network classifier is estimated by testing a total of 600 samples, which are modeled on the basis of the failure model. The average correct classification with and without mind evolution computation algorithm is about 98% and 96.17% for various fault signals respectively. The outcome got from the simulation analysis shows the potency of the proposed hybrid neural network for fault diagnosis in multi phase induction motor.

Deep balanced domain adaptation neural networks for fault diagnosis of planetary gearboxes with limited labeled data

Due to the time-varying working conditions of planetary gearboxes, the labeled data available are usually limited, current intelligent fault diagnosis methods cannot achieve satisfactory results in the case of limited labeled data. In this paper, a novel method called deep balanced domain adaptation neural network (DBDANN) is proposed for fault diagnosis of planetary gearboxes. First, the multiple convolutional layers were used to extract transferable features layer-by-layer from the raw vibration data of source and the target domain. Then, multi-layer balanced domain adaptation is applied to help training the model, which can reduce the discrepancy of the marginal probability distribution and the conditional probability distribution simultaneously. At last, the test data was fed into the model to evaluate the performance. Experiments with multiple cross-domain tasks under varying speeds and loads demonstrate the effectiveness of the DBDANN, and the proposed model obtained a better performance when compared with the state-of-the-art methods.