An adaptive spatiotemporal feature learning approach for fault diagnosis in complex systems

A Multi-Domain Diagnostics Approach for Solenoid Pumps Based on Discriminative Features

“…Step 4: Convert the logarithmic Mel spectrum to timedomain by taking the discrete cosine transform (DCT) of the spectrum as shown in (8).…”

“…On another hand, the issues of sensor data discrepancy (scenarios where testing data distribution is different from the training data) emanating from varying loading, environmental and operating conditions of ICPS components still remain a challenging factor. Although reliable solutions have been proffered in-cluding transfer learning of source domain (training data) with joint distribution adaptation [6], [7], adaptive spatiotemporal feature learning [8], etc., the major concerns with such deep neural processing units range from interpretability, standardized weight initialization paradigm, over-fitting, optimal hyper-parameter selection (and optimization) and in a nutshell, finding the optimal balance between power consumption and performance [2]; however, their efficacy in multimedia data processing (images and videos) speak volumes in industries and academia whereas studies on their effectiveness for real-time monitoring are still ongoing with remarkable progress thus far. Contrary to these deep methods, several M achine learning-based diagnostics tools still retain their robustness and capacity for accurate diagnosis.…”

“…Several vibration-based FD&I techniques have been reported including the use of deep feature learning of spectral images of real-time signals, multi-level decomposition, etc. [1], [8], [11]; however, the importance of feature engineering (feature extraction, selection, reduction/transformation, and normalization) remains one of the major problems faced with the optimum efficiency of these techniques due to the uniqueness in strength (and weakness) of every feature/descriptor for modelling/reflecting a target systems' properties especially from stationary/noisy vibration (or other sensor) signals [11]. Arguably, the use of "every feature" could have dire computational consequences/costs, and can lead to curse of dimensionality, over-fitting and classifier/predictor conf usion.…”

Bearing Intelligent Fault Diagnosis in the Industrial Internet of Things Context: A Lightweight Convolutional Neural Network

“…Traditional mechanical fault diagnosis mainly includes three steps: i) to construct perfect characteristic parameters that can represent bearing faults by using advanced signal processing methods, such as wavelet decomposition, wavelet packet decomposition, empirical mode decomposition, variational mode decomposition, Spectral kurtosis as well as the improved algorithms of the above methods [7]- [9]; ii) to select key feature parameters via dimensionality reduction methods such as principal component analysis, and autoencoder [10]; iii) to realize fault classification through pattern recognition methods, including support vector machine (SVM), decision tree, random forest and artificial neural network methods [11], [12]. However, traditional machine learning methods largely rely on signal processing techniques and diagnostic experience, which makes it difficult to deal with classification or regression problems in complex situations.…”

Learn Generalization Feature via Convolutional Neural Network: A Fault Diagnosis Scheme Toward Unseen Operating Conditions

“…CWRU bearing dataset is used to verify the generalization of the model. Han et al [15]- [18] utilized the spatiotemporal pattern network to explore the relationship among the multiple sensors and diagnose the unseen operating conditions of wind turbine data. Chen et al [19] employed the dep inception net with atrous convolution to extract common features between the artificial data and the natural data in the Paderborn bearing dataset.…”