Machine learning-based anomaly detection via integration of manufacturing, inspection and after-sales service data
| Date | 12 June 2017 |
| DOI | https://doi.org/10.1108/IMDS-06-2016-0195 |
| Pages | 927-945 |
| Published date | 12 June 2017 |
| Author | Taehoon Ko,Je Hyuk Lee,Hyunchang Cho,Sungzoon Cho,Wounjoo Lee,Miji Lee |
| Subject Matter | Information & knowledge management,Information systems,Data management systems,Knowledge management,Knowledge sharing,Management science & operations,Supply chain management,Supply chain information systems,Logistics,Quality management/systems |
Machine learning-based anomaly
detection via integration of
manufacturing, inspection and
after-sales service data
Taehoon Ko, Je Hyuk Lee, Hyunchang Cho and Sungzoon Cho
Seoul National University, Gwanak-gu, Republic of Korea, and
Wounjoo Lee and Miji Lee
Doosan Infracore Co. Ltd, Seoul, Republic of Korea
Abstract
Purpose –Quality management of products is an important part of manufacturing process. One way to
manage and assure product quality is to use machine learning algorithms based on relationship among
various process steps. The purpose of this paper is to integrate manufacturing, inspection and after-sales
service data to make full use of machine learning algorithms for estimating the products’quality in a
supervised fashion. Proposed frameworks and methods are applied to actual data associated with heavy
machinery engines.
Design/methodology/approach –By following Lenzerini’s formula, manufacturing, inspection and after-
sales service data from various sources are integrated. The after-sales service data are used to label each
engine as normal or abnormal. In this study, one-class classification algorithms are used due to class
imbalance problem. To address multi-dimensionality of time series data, the symbolic aggregate
approximation algorithm is used for data segmentation. Then, binary genetic algorithm-based wrapper
approach is applied to segmented data to find the optimal feature subset.
Findings –By employing machine learning-based anomaly detection models, an anomaly score for each
engine is calculated. Experimental results show that the proposed method can detect defective engines with a
high probability before they are shipped.
Originality/value –Through data integration, the actual customer-perceived quality from after-sales
service is linked to data from manufacturing and inspection process. In terms of business application,
data integration and machine learning-based anomaly detection can help manufacturers establish quality
management policies that reflect the actual customer-perceived quality by predicting defective engines.
Keywords Manufacturing process, Machine learning, Data integration, After-sales service,
Anomaly detection, Heavy machinery engine
Paper type Research paper
1. Introduction
In a manufacturing process, it is highly important to manage the quality of the
manufactured product. Poor quality leads to increases in cost of handling defects that occur
within the warranty period, and degradation of reputation in the market. Many
manufacturers collect numerical data from each process equipment in the manufacturing
process, which then serves as the basis of quality control and process management.
Conventionally, statistical process control (SPC) methods have been used to control and
monitor the production process (Shang et al., 2013; Jiang, 2015). Each sensor value derived
from a manufacturing equipment is compared to its upper control limit, lower control limit,
upper inner fence and lower inner fence. Some other SPC methods utilize visualization,
such as control charts, box-plots, stem-and-leaf plots, etc. (MohanaRao et al., 2013).
In inspection steps, the most common method for checking the quality is to select a set of
inspection items in advance and then examine the corresponding values such as
specification error, performance. In the meantime, after-sales service history is another
useful source for maintaining high level of quality. After-sales service data consist of
Industrial Management & Data
Systems
Vol. 117 No. 5, 2017
pp. 927-945
© Emerald PublishingLimited
0263-5577
DOI 10.1108/IMDS-06-2016-0195
Received 2 June 2016
Revised 14 September 2016
Accepted 10 October 2016
The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/0263-5577.htm
927
Machine
learning-based
anomaly
detection
diagnostic information of defects such as symptoms, reasons and corresponding repairs.
Collected data are used to calculate key performance indicators related to quality
management such as defects per unit, problem reporting resolution, worst items.
However, these methods have some limitations. First, limitation comes from
aforementioned SPC methods applied to the manufacturing process. Traditionally,
univariate SPC methods have been widely used because of their intuitiveness
(MohanaRao et al., 2013; Shang et al., 2013). Univariate SPC methods, however, consider
only one variable at a time and cannot capture defects caused by the interaction between
two or more variables. An example is illustrated in Figure 1. Figure 1(a) shows a scatter
plot representing artificial data with two dimensions X
1
,X
2
. There are eight abnormal
points. Suppose that points lying outside of a dotted line for each dimension are predicted
to be abnormal as Figure 1(b). In this case, five abnormal points are classified correctly
while three abnormal points are misclassified as normal. These three abnormal points
existonnormalboundaryineachdimension,thoughtheyarefarawayfromnormal
points in a coordinate plane. Only if an area containing only normal points can be
figured out, as illustrated in Figure 1(c) with a dotted line, all abnormal points can
be classified correctly.
Second limitation is related to the selection process of inspection items. Manufacturers
cannot be sure whether the selected inspection items are sufficient to ensure product quality.
Generally, these items are determined based on current practices and domain knowledge.
X
1
X
2
a normal point
an abnormal point
an abnormal point which is classified
as abnormal
an abnormal point which is classified
as normal
(a)
(c)
X
1
X
2
X
1
X
2
X
1
X
2
(b)
Notes: (a) Scatter plot of artificial data containing eight abnormal points; (b) detecting
abnormal points in each dimension: some abnormal points are misclassified; (c) detecting
abnormal points considering two dimensions: all abnormal points are classified correctly
Figure 1.
Classifying normal
and abnormal points
by univariate and
multivariate methods
928
IMDS
117,5
To continue reading
Request your trial