Machine-based Predictive Maintenance in Manufacturing Industry
Year: 2024
Type: article
Abstract: : Industrial predictive maintenance has changed. We examine the challenges, methods, and results of machine-based predictive maintenance in this paper. Organisations often struggle with infrastructure and data quality. This can include data management issues, technical issues, and data correctness, completeness, and consistency. These issues can lower data-driven process and decision reliability. To maintain their data infrastructure and data quality, organizations must handle these issues. It can be challenging to navigate predictive maintenance. Maintaining data accuracy and reliability is vital. Failure data, which is essential for predictive model training, is often inadequate, non-existent, or manually recorded. Poor sensor data makes it hard to draw conclusions due to the terrain's complexity. Domain experience boosts AI predicting accuracy. Because human skill improves AI, forecasts are more accurate. Generative Data Augmentation creates synthetic data to solve data scarcity. This technique adds data to the dataset instead of using current data. This addresses data scarcity. Generative AI for Operations boosts worker productivity, while Change Management and Monetization ensures change success. These solutions significantly reduce unscheduled downtime and minimize ownership costs for manufacturers. Implementing predictive maintenance outside the shop floor can boost efficiency. Finally, predictive maintenance can increase production and asset management by efficiently navigating the process. Our research is fully described in this booklet, including key findings and producer implications. It stresses data and inventive problem-solving. More research in this area could boost industry performance and modernize maintenance. The data set was thoroughly analyzed, taking into consideration potentially relevant indicators. We evaluated various supervised machine learning techniques, including k-nearest neighbor, Random Forest, and Logistic Regression, through a comparative study. The data set was divided into three parts: training (70%), test (20%), and validation (10%). A cross-validation method was used during modeling, and a testing set was used to validate the prediction performance of the supervised learning models. Model performance evaluation typically involves the use of accuracy measures. Based on the accuracy analysis and correlation matrix, it was found that the k-nearest neighbor classifier outperformed other learning algorithms in predicting the given data set. The optimal hyperparameters for the k-nearest neighbor model were determined to be Manhattan for the metric, uniform for the weights, and 21 for the number of neighbors. Keywords: AI, Machine Learning, Data Quality, Predictive Maintenance, Production Practices, Asset Management, Predictive Models
Authors: Swati Patil, Prajkta Patil, Sonali Sharma, Sudhir Chitnis, Arun Patil
Language: English
Subfield: Safety, Risk, Reliability and Quality
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W6911733313
FWCI:
Cites:
Cited by:
Related to: