Facilitating quality management through data mining

Abstract: In this report, the topics of quality management, knowledge work, and Lean Six Sigma areexplored with the objective of identifying potential improvements that could be facilitated byData mining methods. With the purpose of exploring the topic of knowledge extraction fromfree-text data to support decision-making in manufacturing operations from a qualitymanagement perspective. Due to increased amounts of data being generated by new technologiessuch as Industrial Internet of Things, and increasingly complex production systems, newchallenges are emerging for manufacturing companies (Titmarsh, et al., 2020; Kamm, et al.,2021). These developments can potentially lead to an escalation of the demands on knowledgeworkers within quality management due to additional data processing. Such tasks have beenobserved to potentially increase cognitive load with decreased knowledge work productivity asa consequence, which is a key factor in effective quality management (Yusoff, et al., 2017). Inparticular, free-text data have been observed to be a challenge when implementing data miningmethods.The conducted research was guided by the following three research questions:RQ 1: What are the challenges of manually performed knowledge discovery in data for qualitymanagement?RQ 2: What are the potential benefits of implementing data mining methods for qualitymanagement in Lean Six Sigma companies?RQ 3: How can complex data processing tasks that include free-text analysis be improved?The topics in question were explored by utilizing peer-reviewed literature, which was verifiedby a case study carried out at a Lean Six Sigma industrial equipment manufacturer. The casestudy includes observations and interviews to gain understanding of the case company’soperations within the selected scope. In addition, an experimental solution with the objective ofimproving an inefficient process was developed using data mining methods for free-text analysis,as described in literature. Such methods are commonly referred to as text mining.During the case study, the challenges described in literature related to text mining in general, andwithin the manufacturing industry in particular, were observed and experienced first-hand.Properties such as lack of labelled data samples, imbalanced data sets, insufficient data qualitydue to varying grammar, and the common use of mixed languages make the implementation oftext mining in industrial contexts a challenge (Ittoo, et al., 2016). The achieved results confirmseveral of the challenges described in the reviewed literature. The task of pre-processing data toenable data-driven decision-making at the case company is identified as an inefficient, butnecessary task for extracting useful knowledge from the company’s collected data. In an attemptto identify potential improvements for this particular task, several different experimentalmachine learning models for text classification were developed.While the developed solutions did not show sufficient performance to enable full automation ofthe task, potentially promising solutions were discovered. Particularly the data-driven nature ofthe systematic Lean Six Sigma methodology was observed to be suitable for integrating withdata mining methods for enhanced results (Fahey, et al. 2020; Fahmy, et al. 2017). In addition,reducing non-value adding and cognitively demanding tasks could support in optimizing thecognitive load experienced by knowledge workers and allow for increased performance inquality management activities (Jalani, et al. 2015).