Machine failure prediction using joint reserve intelligence with feature selection technique

Impacts of Feature Selection on Predicting Machine Failures by Machine Learning Algorithms

“…A highlight is the work of [28], who applied joint reserve intelligence and feature selection techniques Classifier Attribute Evaluation (CAE), Correlation Attribute Evaluation (COAE), Infogain Subset Evaluation (ISE), and Classifier Subset Evaluation (CSE) to predict machine failures. This indicates a trend towards incorporating innovative approaches to improve predictive effectiveness.…”

“…The literature on machine failure prediction covers a wide variety of approaches and techniques. Many studies focus on developing machine learning algorithms [9][10][11][24][25][26][27][28][29][30]. Others explore methods of processing signals acquired from machines through sensors [25,31].…”

“…This step involves collecting and pre-processing data. First, a dataset related to predictive maintenance, proposed by [32] and adopted in the study of [28], was collected from the UC Irvine Machine Learning Repository. This dataset has 10,000 records, each composed of the following attributes: UID, a unique identifier ranging from 1 to 10,000; ProductID, which represents the quality of the product, where L represents 50% of all products, M represents 30% of all products, and H represents 20% of all products; air temperature (K), generated using a random walk process subsequently normalized to a standard deviation of 2 K around 300 K; process temperature (K), generated using a random walk process normalized to a standard deviation of 1 K, added to the air temperature plus 10 K; rotation speed (rpm), calculated from the power of 2860 W, overlaid with customarily distributed noise; torque (Nm) values are typically distributed around 40 Nm with σ = 10 Nm and no negative values; tool wear (min), H/M/L quality variants add 5/3/2 min of tool wear to the tool used in the process; target label indicates whether the machine failed or not at this specific data point; and failure type represents a specific type of failure, including no failure, heat dissipation failure, power failure, random failure, overstrain failure, and tool wear failure.…”

“…Since real predictive maintenance data are often difficult to obtain, this dataset represents an important source of data, because according to [28,32], it reflects the real predictive maintenance found in the industry.…”

“…The model training was conducted with 50, 100, and 200 estimators (trees), separating 70% of the data for training and 30% for testing. We used accuracy, precision, and recall metrics to evaluate the model, as indicated in [11,[26][27][28]. Afterward, the Support Vector Machine (SVM) technique was employed.…”