Data-Driven Intelligent Condition Adaptation of Feature Extraction for Bearing Fault Detection Using Deep Responsible Active Learning

Abstract

The detection of faulty bearings is an essential step in guaranteeing the safe and efficient operation of rotating machinery. Bearings, which also transmit the loads and pressures generated by the machinery, support the rotating shafts. A common method for bearing fault diagnostics is using signal processing techniques. In terms of accuracy, dependability, and sensitivity to various fault types and severity levels, these techniques do, however, have significant limits. To address these limitations, practitioners often integrate signal processing with advanced techniques like machine learning and data analytics to enhance diagnostic accuracy, reliability, and overall effectiveness in bearing fault detection and predictive maintenance. Exploration of various models has demonstrated enhanced results in managing nonlinear data to a certain degree; however, these models face challenges when dealing with intricately complex patterns. Moreover, CNNs can automatically learn relevant features from raw sensor data, capturing intricate patterns and relationships in the data without the need for manual feature engineering. CNNs can be optimized for scalability and real-time processing, essential for applications requiring quick decisions and computationally less expensive for large datasets compared to other models. An optimized one-dimensional Convolutional Neural Network (1D CNN) using different kernel sizes is proposed for predicting and finding bearing problems to overcome these constraints. This method creates a feature vector by applying many filters of various sizes to the input signal. Using the created feature vector, the input signal can be divided into many categories, such as healthy or unhealthy. In comparison to other methods, the proposed technology performs better and offers a high accuracy of 99.52% in bearing fault identification. The 1D CNN model with multiple kernel sizes excels in preserving data structure during dimensionality reduction, as confirmed by comparing t-Distributed Stochastic Neighbor Embedding plots. Particularly, the optimized 1D CNN with multiple kernel sizes accurately classifies faults with minimal errors, showcasing its fault classification proficiency compared with the other state-of-the-art methods. The visualization underscores the methodology's efficacy in discerning intricate fault patterns within the data.

Advanced AI-driven approach for enhanced brain tumor detection from MRI images utilizing EfficientNetB2 with equalization and homomorphic filtering

“…A significant technological gap persists in brain tumor detection using machine learning (ML) and artificial intelligence (AI), primarily due to limited access to large and diverse datasets for training robust models, challenges in interpreting complex ML algorithms, and concerns regarding model generalizability across different populations and healthcare settings [ 10 , 11 ]. Closing this gap requires collaborative efforts to curate comprehensive datasets, develop explainable AI techniques, and establish rigorous validation protocols [ 12 ]. Addressing these challenges will unlock the full potential of ML and AI in revolutionizing brain tumor detection and enhancing patient outcomes and to solve this only this research is proposed.…”

Detection and Classification of Rolling Bearing Defects Using Direct Signal Processing with Deep Convolutional Neural Network

“…One solution with good results was the use of deep neural networks [10,[34][35][36][37]. The structure of deep neural networks is based on the use of many different neural layers [15], each of which introduces a certain level of abstraction by extracting successive features from the data coming from the previous layer, starting from the input layer [38,39]. This also makes use of features extracted from processed diagnostic signals as an input for deep network work [10,40]; however, an important advantage of deep networks is their ability to extract features directly from the diagnostic signal and their ability to process them for appropriate classification [38,39,41,42].…”

“…The most widely used deep learning network structures are autoencoders [43], long short-term memory (LSTM) [7], and convolutional neural networks (CNNs) [36,[44][45][46]. Compared to the shallow structures mentioned above, CNNs show higher accuracy when operating directly on the diagnostic signal presented in the form of multidimensional arrays or vectors [38,45], further reducing the time of the diagnostic process [36,37]. CNNs also provide automatic symptom extraction, reducing the role of an expert in the diagnostic process.…”

“…In the approach proposed in this article, a structure with 64 thousand parameters is used. In addition, a dropout layer [10,38] and stochastic network training methods were used to avoid a loss of generalization ability.…”