China
Africa
Explore chapters and articles related to this topic
Instrumentation and Control Systems in Ring and Rotor Spinning
Published in L. Ashok Kumar, M. Senthilkumar, Automation in Textile Machinery, 2018
L. Ashok Kumar, M. Senthilkumar
With the aid of the spectrogram, periodic faults can be detected very easily. At the same time, it helps to judge the sliver and thus the previous working stages. The controllable range of wavelengths extends from 2 cm up to a maximum of 300 m. When exceeding a selectable limit, it is possible to issue a spontaneous report and/or to block the spindle. The asterisks depicted denote the mean values of the machine as a whole. The crosses indicate that the machine mean value and the spindle value are identical. A total of four different spectrograms can be displayed and printed out: Difference spectrogramSpindle spectrogram, last calculated spectrogramSpindle mean value spectrogram, mean value of a spindleMachine spectrogram, mean value of the machine
Signal Analysis
Published in Russell L. Herman, An Introduction to Fourier Analysis, 2016
The spectrogram is created using what is called the "Short-Time Fourier Transform," or STFT. This function divides a long signal into smaller blocks, or windows, and then computes the Fourier transform on each block. This allows one to track the changes in the spectrum content over time. In Figure 7.12 one can see three different blobs in the 3 kHz-4 kHz range at different times, indicating how the three chirps of the bird can be picked up. This gives more information than a Fourier analysis over the entire record length.
Electromyograms
Published in A. Bakiya, K. Kamalanand, R. L. J. De Britto, Mechano-Electric Correlations in the Human Physiological System, 2021
A. Bakiya, K. Kamalanand, R. L. J. De Britto
A spectrogram is a time-frequency representation of a one-dimensional signal, which is an extension of the fast-Fourier transform. In general, the spectrograms provide spectral features in time domain and the distribution of energy in time-frequency domain.
Eigen-spectrograms: An interpretable feature space for bearing fault diagnosis based on artificial intelligence and image processing
Published in Mechanics of Advanced Materials and Structures, 2023
Eugenio Brusa, Cristiana Delprete, Luigi Gianpio Di Maggio
This study proposes a methodology for bearing fault diagnosis, based on spectrogram image processing. The proposed approach (Figure 1) is able to accurately detect bearing faults and classify their type and severity by means of AI spectrogram recognition, also in presence of noisy data. Furthermore, this paper presents a reading key to link features contribution to model results, that is interpretability [67]. One of the core parts is embodied by the implementation of the so-called “Randomized Linear Algebra” (RLA) for intelligent recognition of signal processing outcomes such as spectrograms. RLA is acknowledged as one of the cornerstones of modern data science since it represents an extremely streamlined data mining tool for extracting dominant low-rank structures underlying big datasets [68]. Given that these are expected to populate machinery IFD in the upcoming IoT era [22–24], Randomized Algebra may likewise arise as a groundbreaking engineering tool.
In-process monitoring of the ultraprecision machining process with convolution neural networks
Published in International Journal of Computer Integrated Manufacturing, 2023
K Manjunath, Suman Tewary, Neha Khatri, Kai Cheng
While analyzing the accelerometer signals in the feed direction, it is better to consider the periodic components in frequency and change in the time domain (Cheng et al. 2015). The drawbacks, like subtle changes in signals, are difficult to notice in a time domain. Whereas, if features spread in a wide range of the spectrum, it is difficult to use the Fourier Transform (FT). FT reveals the signals’ lumped information but does not show the signals’ temporal structure. Further, a spectrogram of it is generally created from a time signal using the Fast Fourier transform (FFT).
Indian language identification using time-frequency image textural descriptors and GWO-based feature selection
Published in Journal of Experimental & Theoretical Artificial Intelligence, 2020
Amit A. Chowdhury, Vaibhav S. Borkar, Gajanan K. Birajdar
Figure 1 depicts the architecture of the proposed Indian language identification algorithm using GWO feature selection and ANN classifier. From Figure 1, the training phase consists of four steps. Six Indian languages from Indic TTM-Database TTS Database (2017) are used for training and testing purposes. The input speech samples are in wav format. In the first stage, we convert these audio samples into a spectrogram image. Spectrogram converts the signal into a time-frequency-based representation. In the second stage, three different tetxure descriptors including LBPHF, CLBP and Wavelet are used to extract the features from the spectrogram image. LBPHF feature extraction step results in a feature matrix of dimension for each language, where is the total no. of samples and is LBPHF feature vector dimension. For CLBP, the dimension of the feature matrix is where is CLBP dimension and for wavelet, it is ( shows the length of DWT features). These three texture features are combined finally with dimensions before applying to GWO. In the third stage using grey wolf optimizer (GWO), the optimal feature set is selected from the extracted features. It is a process of selecting a significant subset of features to create a more accurate model. And finally, these selected features are used to train the neural network classifier. A feed-forward neural network as a classification model is employed to observe some hidden pattern in the feature matrices which will help us to recognise the languages of the unknown samples.