signal processing

In recent decades, data-driven methods have shown high effectiveness in structural health monitoring, focusing on damage detection. In the past, using raw time-series sensor signals for diagnosing structures was nearly impossible due to the inherent high dimensions of the measured data. However, recent advances in neural networks and machine learning models have eliminated the need for complex preprocessing of time-series data. Monitoring based on access to healthy and damaged structural information in a classification framework has been the predominant approach in recent research .This study conducts a comparative analysis of four novel deep-learning algorithms for structural damage detection using raw data within a supervised monitoring framework. The four algorithms investigated include LSTM networks, GRU, CNN, and a hybrid CNNLSTM network. A precise approach (the CNNLSTM model with 99.9% accuracy) is proposed based on the Los Alamos Laboratory foundational structure data. Notably, the CNNLSTM model operates without any preprocessing, such as signal processing algorithms or dimensionality reduction, and performs excellently in terms of accuracy, the number of parameters, and training time compared to the other three models and previous studies.

Fault diagnosis of mechanical systems is of special importance for better system performance as well as its protection. In this work, a rotary motor laboratory system is used to generate signals and data. The obtained data are placed in the pre-processing process. In this article, to improve the performance of signal analysis, the combined analysis methods using signal features and Kalman filter are proposed. At first, the Kalman filter is used to reduce the signal noise. And in the following, for signal pre-processing, the features of the signal in the time domain and frequency domain are suggested, which have been used as one-dimensional signal pre-processing. In the following, several neural networks such as support vector machine, multilayer perceptron, and convolutional neural networks have been used to analyze the obtained features. To check the results, the data is divided into training data and validation data. Accuracy results for validation data are examined in different methods. The results indicate the better performance of the AlexNet convolutional neural network in the presence of the Kalman filter denoising process. In this case, this network has reached an average of 96.1% accuracy for validation data, which has been improved compared to other classifiers and fault diagnosis without noise reduction.

In this study, the application of wavelet transform in the natural degradation of roller bearings based on the processing of acoustic emission signals has been investigated. In order to extract the bearing acoustic emission signals, a suitable laboratory arrangement and data collection equipment was used. The processing of acoustic emission signals in time-frequency domains was performed using wavelet transform analysis. In order to capturing the acoustic emission signals, three sensors namely Pico, R6α and WSα were applied. The results showed that the acoustic emission method is an efficient method for estimating the bearing natural degradation. In this study, to select the mother wavelet, the method of comparing the ratio of energy to converted signal entropy was used. Comparison of energy versus entropy for the 53 mother wavelets studied shows that the Bior 3.1 mother wavelet has the highest energy to Shannon entropy ratio and is selected as the best mother wavelet. Also, the performance analysis of the three sensors showed that all three sensors used are suitable for investigating the natural degradation of the bearing, however, the Pico sensor scored higher.

In this study, a new method for the fault detection of the locomotive engine injector nozzle based on vibration analysis and statistical tests, inside artificial neural networks, is presented. For this point, first the under study received vibration signals in the frequency domain is divided into several smaller ranges and the RMS of each range is then extracted as a frequency property and given as an input to the neural network. Because the high selection of the features reduces the accuracy of the neural network, the extracted feature vector with different levels of significance passes through the T-test filters, firstly, and then enters the neural network as an input. Using of this method, the accuracy of the neural network increases from 78.4 to 94.6%, and also help to detect the frequency ranges. According to the results, the fault of the injector nozzle crack increases the intensity of vibrations in the upper band frequencies of 1500 Hz.

Structures suffer local damages due to various reasons. If these damages are not identified, they could be aggravated by natural disasters such as earthquakes or artificial factors such as unprincipled excavations, leading to total destruction of the structure. As a result, health monitoring in structures and their elements is regarded as one of the most significant research topics in civil, mechanical, and aerospace engineering. A damage detection method is to process the time or frequency domain of structural responses. In this context, wavelet transform is one of the processing methods for both time and frequency domains, and many studies have utilized it to investigate structural health monitoring. In this paper, a comprehensive review was conducted on the published studies. Besides, damage detection was performed by considering the vibration mode shapes of a defective cantilever, which were used as wavelet-transform input signals. The diagram of the detail coefficients resulting from the wavelet analysis of the input signal presented the maximum and minimum values in the defective positions. These findings support the efficiency of using wavelet transform in damage detection. Further, the sensitivity of damage detection based on wavelet transform to the severity of damages showed that with the rise in damage severity in a location, a larger relative jump occurs in the output signal of all modes.

Vibration analysis is one of the most practical methods for monitoring and troubleshooting rotating equipment. In this research, vibration analysis and support vector machine algorithms were used for monitoring and troubleshooting Alstom locomotive blowers. First, vibration data were collected from the blowers and the received signals were categorized into four groups: healthy blowers and blowers with problems of unbalance, loose shaft (base), and warped blades. Sixteen frequency and time features were then extracted from the received signals. Because in rotating systems, the ratio of the intensity of vibrations in the harmonics of the rotation of the machine can help diagnose the faults, the ratios of all features were calculated and defined as new features. The accuracy of the network can be sometimes lowered by the multitude of features, thus, a t-test filter was inserted into the support vector machine algorithm to select the features. The results show that the t-test filter increased the accuracy of the support vector machine algorithm. Finally, the feature selection of this network was compared with the feature selection by the genetic algorithm. The results show that the network designed in this research has a better performance in feature selection than the genetic algorithm.

The helical gear systems have some distinctions such as more precision, long life and more applications in the industry, and less vibration, noise and transmission error compared to spur gear systems. One of the damages that can affect the operation of these systems is the gear tooth root crack. So, the crack detection in its early stages of growth and development is very important. The residual signal method is one of the common methods to extract the effect of crack from the system vibration signal. Since the residual signal calculation requires the information of the vibration response of the healthy state or the system initial parameters, providing a method to extract the effect of crack from the vibrating response without any more information is very helpful. In this research, at first the calculation of the helical gear pair mesh stiffness is explained, and the effect of tooth root crack is also described. Then, a single-stage helical gear system, with the motor and the load, is modeled and solved numerically to achieve the system dynamic response. Then, the effect of present method on the vibration signal obtained from the dynamic simulation of a system with a small crack is Shown and discussed. Finally, the proposed method is applied to the dynamic response signal of a system that presented in one of the previous researches, and its efficiency is illustrated.

In this paper, a new active vibration control system has been proposed for the elimination of boring bar chatter in the internal turning process. The system is composed of a boring bar equipped with electromagnetic actuator and accelerometer, as well as a novel adaptive control algorithm that is widely used in the field of active noise control. The controller is known as feedback FxNLMS and is composed of two finite impulse response adaptive filters. One of the filters is known as a model filter, which predicts the dynamic model of actuator-boring bar assembly. The other is known as the control filter and anticipates the inverse model of forwarding path dynamics. The weight vector of the adaptive filter is adjusted by using the normalized least mean square algorithm. Firstly, the impact test is conducted in the presence of an adaptive controller. It is observed that the magnitude of the dominant mode on the forward path’s frequency response function is drastically suppressed by 36 dBs. Secondly, the internal turning tests are conducted on Aluminum alloy 6063-T6, to investigate the performance of the adaptive controller for the purpose of chatter mitigation. Due to the optimal performance of the adaptive controller, the dominant magnitude of the boring bar’s power spectral density is successfully attenuated up to 68 dBs, and the critical limiting depth of cut is increased by 10 folds. Also, the roughness of the machined surface is remarkably improved by 8 folds compared to the control-off cutting test. Moreover, the actuator cost is considerably reduced by 3 folds in comparison to the optimal constant-gain integral controller.

Although clearance in mechanisms facilitates manufacturing and assembly, it will lead to undesirable results such as an increase in vibration and a decrease in accuracy. Joint clearance is considered as the main cause of errors in the position and the direction of the links. This research has investigated the possibility of using encoder signals to estimate the extent of clearance in the joints of a slide crank servomechanism. The output signal of the encoder was studied in three modes of zero, low and high clearances. It was done on the point of the part that serves as a link for servomechanism. The output signals of encoder were tested, analyzed and compared with those in joints with/without clearances. It was found that the signals had a sinusoidal waveform in the joint with ideal clearance but it has a distorted waveform in the joint which has high clearance. The results show that the distortion available in encoder signals may be used as a criterion to estimate and monitor clearance condition in slide crank mechanisms