Product failure prediction with missing data using graph neural networks
- PDF / 506,954 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 10 Downloads / 245 Views
(0123456789().,-volV)(0123456789(). ,- volV)
ORIGINAL ARTICLE
Product failure prediction with missing data using graph neural networks Seokho Kang1 Received: 14 May 2020 / Accepted: 27 October 2020 Ó Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract In real-world production data, missing values often occur randomly or systematically with various missing patterns. Missing values need to be handled properly to build effective prediction models. This paper presents a novel method based on graph representation and graph neural networks for improving prediction in missing value conditions. To utilize the entire information of a training dataset without direct manipulation, all instances of the dataset are represented as graphs of varying sizes, in which nodes and edges represent the observed input variables and their pairwise relationships. Prediction models learn from the graph representations. These models can make predictions of unknown labels for new instances that have arbitrary missing patterns. The superiority of the proposed method was investigated on seven different product failure prediction tasks from a home appliance manufacturer. The proposed method outperformed all other methods in six of the seven tasks. Keywords Failure prediction Production data Missing value Graph neural network
1 Introduction Large volumes of production data are generated and collected by modern manufacturing systems [26, 32]. Through advances in machine learning, these data have become useful sources for data-driven prediction of product failures [6, 19, 35]. The predictive modeling task can be formulated such that each product is an instance, whose input variables are quality-related production factors, such as process parameters, measurements, and inspection results, and whose output variables indicate the occurrence of product failures. Prediction models learn from data for previously manufactured products to predict the output variables. By successfully predicting failures using the models, defective products can be effectively filtered out before shipment to the market. This helps manufacturers to perform preventative maintenance to avoid additional costly processing of defective products. The modeling can contribute to identify & Seokho Kang [email protected] 1
Department of Industrial Engineering, Sungkyunkwan University, 2066 Seobu-ro, Jangan-gu, Suwon 16419, Republic of Korea
root causes of failures, thereby improving the quality of future manufactured products [4]. It can also be extended to investigate on the optimal process parameters for enhanced process control within the manufacturing system. The success of the predictive modeling task depends largely upon the quality of the production data used. An important consideration regarding the quality is that data are assumed complete. Ideally, all variables for every instance are filled with observed values [8, 30]. However, in real-world situations, the values of some variables may not be observed depending on the nature of the data collection env
Data Loading...
