Adaptive SVM-based real-time quality assessment for primer-sealer dispensing process of sunroof assembly line

https://doi.org/10.1016/j.ress.2018.03.020Get rights and content

Highlights

  • A real-time quality monitoring process using an adaptive SVM algorithm is proposed.

  • The quality assessment performance can improve over time by A-SVM.

  • The proposed system is verified and validated in the primer sealer manufacturing.

  • The performance of A-SVM is better than other machine-learning methods.

Abstract

Quality assessment in many production processes typically relies on manual inspections due to a lack of reference data and an effective method to classify defects in a systematic way. Recently, the real-time, automated approach for product quality assessment has been regarded an important aspect for smart manufacturing applications, such as in the automotive industry. In this research, we suggest a framework to pre-process the data for SVM-based decision making and implement the algorithm in the self-evolving quality assessment system based on the adaptive support vector machine (ASVM) model. An adaptive process is a feedback control that ensures the effectiveness of the support vector machine (SVM) algorithm over time and enables the improvement of SVM-based quality assessment in the real production process. Next, an industrial case study of a primer-sealer dispensing process in a sunroof assembly line of an automobile is illustrated with statistical analysis to verify and validate the applicability and effectiveness of the proposed ASVM-based quality assessment system. Defective patterns are then analyzed using an infrared thermal image of primer-sealer dispensing in a manufacturing process, which contains multi-modal data of dimensional information and temperature deviation from the dispending patterns in our study.

Introduction

Recent manufacturing companies have faced challenges to manage the manufacturing complexity due to an increasing requirement of product variety from the customers [1]. Thus, the automated system is an alternative to enhance productivity and improve cost efficiency in the global, complex competition in many firms. In addition, with an advance in sensor technologies and computer capabilities, the real-time, data-driven tools have been adapted to examine products’ quality in many industrial cases [2]. Several industries adopt sensor and camera technologies in their process lines to improve safety conditions for operators and making the process more reliable [2], [3], [4], [5], [6]. An aid of real-time data acquisition also enables an application of machine learning algorithms to accurately assess quality of manufacturing process. Based on the concept of supervised learning, the machine learning techniques are operated by building a model from a training set of input observations in order to make data-driven predictions or decisions expressed as outputs [7], [8], [9]. Some of these algorithms are artificial neural network, support vector machine (SVM), and Bayesian statistic.

Machine learning techniques have been lately implemented in the quality inspection to predict and reduce errors in the classification results based on a variety of data analysis methods (e.g., see [10], [11]). The SVM algorithm, in particular, has been shown to be an effective computational classification method widely used in many engineering applications (e.g., [11], [12]). Researchers have also proposed to improve the SVM algorithm through an analysis of optimal parameters for ensuring the applicability of the method (e.g., [13], [14]). Although previous studies with regard to SVM applications have been proposed in industries, an implementation in a real manufacturing study with long-term production data is scarcely reported due partly to the sensitivity of manufacturing and operational issues. A feedback (adaptive and self-updating) process with a large amount of historical quality data is important to ensure the effectiveness of the SVM algorithms over time, which enables the real-time improvement of SVM-based quality inspection [15], [16].

In this study, we propose the adaptive SVM (ASVM) with quality manager's judgement. The real-time quality assessment using ASVM is implemented in a production line on a daily basis so that the proposed algorithm will systematically repeat the SVM-training loop when the system judgement of the manufacturing process is found to be inconsistent with human/expert judgement. This closed-loop process enhances the system's accuracy over time and ensures reliability in the long run. The proposed system is then verified and validated by evaluating the quality of the primer-sealer dispensing process of an automotive assembly line. The primer-sealer dispensing process is a process that a frame and a glass are attached though the heating process in an automotive company. Thus, an analysis of the thermal image data based on the infrared thermography (IRT) is needed and further conducted to assess the quality information from the heating process that cannot be easily assessed by a human operator. The thermal image data is collected, pre-processed, and analyzed using image processing algorithms in our study. In particular, the key novelty of this study is emphasized in three folds: Frist, we pre-process and integrate the IRT data for machine learning-based analysis using ASVM to assess the quality information related to temperature profile of the automotive manufacturing process. Second, the adaptive process in the ASVM-based quality classification is an important part of the proposed system found to effectively enhance the system's accuracy and reliability over time. Third, the actual implementation is illustrated, in which the part quality is classified as either defective or good to be displayed real time on the kiosk personal computer (PC), and is adopted as a part of the company quality inspection system.

The remaining sections of this paper are organized as follows. We introduce the relevant studies in Section 2. Next, we discuss the real-time quality assessment using the ASVM algorithm in Section 3. Then, the case study in an automotive industry using primer-sealer dispensing process and thermal image data analysis is illustrated in Section 4. Finally, we present our research conclusions and outline future research directions in Section 5.

Section snippets

Literature review

Machine learning and computational statistics (e.g., statistical process control (SPC) are closely related fields from methodological principles to theoretical tools, in which the former evolves from the study of pattern recognition and computational learning theory in artificial intelligence and expert system. In particular, machine learning approach explores the construction of algorithms that can learn from predictions on data by building a model from a training set of input observations to

Problem statement

Generally, product quality has been maintained by skilled operators on the basis of their experience and intuition [41]. However, not only are the manual inspection system slow, but also ineffective. In fact, some research has stated that human inspectors typically find about 80% of the defects [42]. Meanwhile, with the advances in the sensing technology, several manufacturers have shifted from manual to automated quality inspection. Ironically, good number of studies argues that fully

Case study and application

This section discusses the ASVM-based quality assessment using a real case study data from the primer-sealer dispensing process. We first introduce the sunroof assembly process for an automotive requiring the primer-sealer dispensing process. Next, we discuss image data preparation. Then, we compare our ASVM-based results with the results obtained from an expert judgment alone. Finally, we discuss the key findings from the case study.

Conclusions and future research

We propose the ASVM-based quality assessment using thermal image data in the primer-sealer dispensing process in this study. The proposed quality assessment system incorporating human/expert judgement is implemented and applied to enhance the system accuracy over time in the long run. The proposed system consists of the following key operations. First, a number of thermal cameras with their respective locations and timing were determined based on the product and process being examined. Then,

Acknowledgments

This research was supported by the 5th regional S/W convergence business though the Ulsan Business Support Center and the National IT Industry Promotion Agency (NIPA) funded by The Ministry of Science, ICT, Future Planning (MSIP), and Ulsan metropolitan city, Republic of Korea. The authors would like to also thank the three anonymous reviewers for comments, which greatly improve the quality of the manuscript and its presentation.

References (57)

  • Y-H Liu et al.

    Automatic target defect identification for TFT-LCD array process inspection using kernel FCM-based fuzzy SVDD ensemble

    Expert Syst Appl

    (2009)
  • Z Xue-Wu et al.

    A vision inspection system for the surface defects of strongly reflected metal based on multi-class SVM

    Expert Syst Appl

    (2011)
  • A Widodo et al.

    Combination of independent component analysis and support vector machines for intelligent faults diagnosis of induction motors

    Expert Syst Appl

    (2007)
  • A Lebar et al.

    Method for online quality monitoring of AWJ cutting by infrared thermography

    CIRP J Manuf Sci Technol

    (2010)
  • M Kano et al.

    Data-based process monitoring, process control, and quality improvement: recent developments and applications in steel industry

    Comput Chem Eng

    (2008)
  • MW Gardner et al.

    Artificial neural networks (the multilayer perceptron)—a review of applications in the atmospheric sciences

    Atmos Environ

    (1998)
  • M Busogi et al.

    Computational modelling of manufacturing choice complexity in a mixed-model assembly line

    Int J Prod Res

    (2017)
  • C Meola et al.

    Recent advances in the use of infrared thermography

    Meas Sci Technol

    (2004)
  • Venkatraman B, Menaka M, Vasudevan M, Raj B. Thermography for online detection of incomplete penetration and...
  • L Serio et al.

    Monitoring of the friction stir welding process by means of thermography

    Nondestr Test Eval

    (2016)
  • C Bishop

    Pattern recognition and machine learning (Information science and statistics)

    (2007)
  • P Langley

    The changing science of machine learning

    Mach Learn

    (2011)
  • E. Alpaydin

    Introduction to machine learning

    (2014)
  • M das Chagas Moura et al.

    Failure and reliability prediction by support vector machines regression of time series data

    Reliab Eng Syst Saf

    (2011)
  • W Li et al.

    Design of online monitoring and fault diagnosis system for belt conveyors based on wavelet packet decomposition and support vector machine

    Adv Mech Eng

    (2013)
  • G-M Lim et al.

    Fault diagnosis of rotating machine by thermography method on support vector machine

    J Mech Sci Technol

    (2014)
  • D Tuia et al.

    A survey of active learning algorithms for supervised remote sensing image classification

    IEEE J Sel Top Signal Process

    (2011)
  • J Harding et al.

    Data mining in manufacturing: a review

    J Manuf Sci Eng

    (2006)
  • Cited by (26)

    • Evaluation of dynamic stress level and physiological change for live salmon in waterless and low-temperature transportation

      2021, Aquaculture
      Citation Excerpt :

      We incorporate an IoT and AI-model to illustrate the coupling relationship of various quality indicators monitoring (mainly micro-ambient and stress level indicators). The objective is to build the biosensing strategy, and achieve stress change modeling based on obtained data by smart sensing devices (Oh et al., 2018; Arsalane et al., 2018; Tanajura et al., 2015; Sun et al., 2019). Multi-sensors ambient indicators and breath rate stress indicator monitoring system were designed and implemented.

    • Process monitoring for quality — A multiple classifier system for highly unbalanced data

      2021, Heliyon
      Citation Excerpt :

      These algorithms allow for people to produce reliable, repeat-able decisions and uncover hidden insights by learning from historical relationships and trends in the data. An algorithm integrating feedback and an adaptable process that is analyzed over time in the long run to improve SVM-based systems was proposed in [27]. The adaptive approach considerably improved the feedback process and effectively enhanced the system accuracy and reliability over time.

    • A manufacturing quality prediction model based on AdaBoost-LSTM with rough knowledge

      2021, Computers and Industrial Engineering
      Citation Excerpt :

      It is noted that all the SVM, the FFNN and the ANFIS are the typical representations of shallow learning, that is, the SVM uses kernel function for regression (none hidden layer), the traditional FFNN has one hidden layer, and the ANFIS is a framework with a fuzzy rule function. The details of the three methods can be found in references (Oh et al., 2019; Mahamad et al., 2010; Cheng et al., 2018). The LSTM with rough knowledge and the AdaBoost-LSTM are used to verify the influence of reinforcement strategy and knowledge extraction on the single LSTM model improvement.

    View all citing articles on Scopus
    View full text