Proceedings of the
9th International Conference of Asian Society for Precision Engineering and Nanotechnology (ASPEN2022)
15 – 18 November 2022, Singapore
doi:10.3850/978-981-18-6021-8_OR-08-0094
Few-Shot Learning for Prescriptive Maintenance
1Singapore Institute of Manufacturing Technology (SIMTech) @ Fusionopolis 22 Fusionopolis Way #08-04, Innovis Singapore 138634
ABSTRACT
Machine learning has been applied to various areas, but one limitation of the industry to use machine learning algorithms is that they do not have enough data. That is, most machine learning algorithms are still data-driven. Few-shot learning is a type of machine learning algorithm that focuses on processing the dataset containing few samples. Many different areas, such as medical, are exploring few-shot learning due to the sensitivity of their data. The emergence of few-shot learning allows us to use the few data for data analytics.
Collecting data for the industry takes a long time because error cases are rare in daily operations. So, the industry takes one or two years to collect data, and then they can start data analytics. With few-shot learning, the time cycle can be shortened to some extent. The model-agnostic meta-learning (MAML) is one of the most popular few-shot learning algorithms widely used nowadays. Thus, we customize the MAML algorithm and apply it to our prescriptive maintenance tasks to help improve the accuracy in predicting the failure when there are only hundreds of data samples. The experimental results on the real-world prescriptive maintenance dataset show that our customized algorithm achieves higher accuracy than the traditional algorithm.
Keywords: Few-Shot Learning, Meta--Learning, Prescriptive Maintenance
1Singapore Institute of Manufacturing Technology (SIMTech) @ Fusionopolis 22 Fusionopolis Way #08-04, Innovis Singapore 138634
ABSTRACT
Machine learning has been applied to various areas, but one limitation of the industry to use machine learning algorithms is that they do not have enough data. That is, most machine learning algorithms are still data-driven. Few-shot learning is a type of machine learning algorithm that focuses on processing the dataset containing few samples. Many different areas, such as medical, are exploring few-shot learning due to the sensitivity of their data. The emergence of few-shot learning allows us to use the few data for data analytics.
Collecting data for the industry takes a long time because error cases are rare in daily operations. So, the industry takes one or two years to collect data, and then they can start data analytics. With few-shot learning, the time cycle can be shortened to some extent. The model-agnostic meta-learning (MAML) is one of the most popular few-shot learning algorithms widely used nowadays. Thus, we customize the MAML algorithm and apply it to our prescriptive maintenance tasks to help improve the accuracy in predicting the failure when there are only hundreds of data samples. The experimental results on the real-world prescriptive maintenance dataset show that our customized algorithm achieves higher accuracy than the traditional algorithm.
Keywords: Few-Shot Learning, Meta--Learning, Prescriptive Maintenance
