Machine learning (ML) algorithms for detecting defects and predictive maintenance of industrial equipment have emerged as a set of critical methods for improving operational efficiency, reducing unexpected downtime, and extending machinery life. This study presents a comprehensive examination of various machine learning models, signal processing approaches, and reduced dimensionality methods for monitoring system health and detecting possible flaws based on research published in the last five years. Convolutional Neural Networks (CNN), Support Vector Machines (SVM), Genetic Algorithms (GA), Multi-Layer Perceptron (MLP), and Fully Connected Neural Networks (FCNN) are utilized for classifying fault patterns coming from sensor data. In contrast, signal processing techniques such as Mel-Frequency Cepstral Coefficients (MFCC) and Short-Time Fourier Transform (STFT) are used to collect significant features from vibration and acoustic signals. Dimensionality reduction approaches such as Principal Component Analysis (PCA), t-distributed Stochastic Neighbour Embedding (t-SNE), ISOMAP, Independent Component Analysis (ICA), and Autoencoders (AE) are used to simplify complex data structures and show crucial defect signals. Random Forest, K-Nearest Neighbours (KNN), and CatBoost are some of the algorithmic ensembles learning methods studied for prediction accuracy and robustness. Furthermore, advanced deep learning models, such as 1D Deep Convolutional Neural Networks (1D-DCNN) and ResNet-3N, are utilized to capture temporal and spatial patterns in time-series data, leading to a more complete comprehension of fault dynamics. The research shows the effectiveness of these various approaches in boosting fault detection systems and improving maintenance techniques, paving the way for intelligent technologies in modern manufacturing.
Abdullah et al. (Thu,) studied this question.