Application of association analysis and natural language processing to improve maintenance management

Abstract

With the advancement of technology in various industrial sectors, companies have been generating large amounts of data at all times. These data not only reveal a company's history, but hide relevant patterns that, if strategically explored, can give the company competitive advantages. For this issue, Data Science has stood out as a science that brings effective solutions through a wide variety of techniques that not only clean, structure and extract information from databases, but also provide useful information/indicators for decision-making processes. In the maintenance management field, the company’s failure report database represents an important asset, but has been little explored regarding their existing failure patterns and relationships, which may provide important improvements to the maintenance management systems. The Association Analysis is a sophisticated Data Science technique used to identify cause-and effect relationships among item sets of the most diverse nature, like code numbers and words. Also, Natural Language Processing is a set of Data Science techniques that support the textual data processing to overcome all the language challenges faced when managing this type of data, and provide relevant portions of it to be explored. The process of extracting knowledge from databases is called Knowledge Discovery in Database (KDD) and this process aims, not only to extract relevant information from databases, but also to support decision-making processes. This research aims to propose and apply a KDD Process, which unifies Natural Language Processing techniques with Association Analysis to process a failure report database, and out of its results, imply maintenance management improvements. The KDD Process’ output in the application section revealed the existence of relevant patterns and strong cause-effect relationships among sets of failure codes and among sets of words presented in the failure descriptions. The knowledge obtained in those files was committed to relevant improvements in different maintenance management processes, like scheduling, team assignment, spare-parts replenishment, resource distribution, FMEA/FMECA/RCM, and so on.