Literal-Aware Knowledge Graph Embedding for Welding Quality Monitoring: A Bosch Case

Zhipeng Tan,Baifan Zhou,Zhuoxun Zheng,Ognjen Savkovic,Ziqi Huang,Irlan-Grangel Gonzalez,Ahmet Soylu,Evgeny Kharlamov

2023-08-02

Abstract:Recently there has been a series of studies in knowledge graph embedding (KGE), which attempts to learn the embeddings of the entities and relations as numerical vectors and mathematical mappings via machine learning (ML). However, there has been limited research that applies KGE for industrial problems in manufacturing. This paper investigates whether and to what extent KGE can be used for an important problem: quality monitoring for welding in manufacturing industry, which is an impactful process accounting for production of millions of cars annually. The work is in line with Bosch research of data-driven solutions that intends to replace the traditional way of destroying cars, which is extremely costly and produces waste. The paper tackles two very challenging questions simultaneously: how large the welding spot diameter is; and to which car body the welded spot belongs to. The problem setting is difficult for traditional ML because there exist a high number of car bodies that should be assigned as class labels. We formulate the problem as link prediction, and experimented popular KGE methods on real industry data, with consideration of literals. Our results reveal both limitations and promising aspects of adapted KGE methods.

Artificial Intelligence

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the welding quality monitoring problem in the manufacturing industry, especially in automobile manufacturing. Specifically, the paper focuses on two core issues:

1. **The size of the welding point diameter**: The welding point diameter is a key indicator for judging welding quality. It must exceed a certain threshold, because a too - small diameter means insufficient connection and may lead to serious consequences (such as the safety problems of automobile users). At the same time, the diameter cannot be too large either, because this will lead to energy waste and may affect the quality of surrounding welding points (for example, through the short - circuit effect).

2. **Which body part the welding point belongs to**: Understanding the proportion of good welding points in each body part is crucial for ensuring product quality. According to the quality standard, this proportion must be higher than a specific threshold.

These two problems are difficult to handle with traditional machine - learning methods, because there are a large number of body parts that need to be used as classification labels, which increases the complexity of the problem. The paper proposes a method based on Knowledge Graph Embedding (KGE) to solve these problems, aiming to improve the efficiency and accuracy of welding quality monitoring and reduce the dependence on traditional destructive testing methods, which are not only costly but also waste - generating. By transforming the problem into a link prediction problem and experimenting with multiple popular KGE methods on real industrial data, the research team explored the potential and limitations of KGE in industrial applications.