Improving automated machines reliability with data analysis
Matikainen, Henri (2018)
2018
Teknis-luonnontieteellinen
Teknis-luonnontieteellinen tiedekunta - Faculty of Natural Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2018-12-05
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tty-201810242448
https://urn.fi/URN:NBN:fi:tty-201810242448
Tiivistelmä
In digital era, massive amount of data is gathered from machinery. This data can be used in multiple ways, but conducting suitable analysis is often not done extensively by companies. This thesis aims to use data analysis as a tool to use gathered data from an operational container terminal to improve machines, especially automated Straddle Carriers, reliability. Machine learning is used to create models, which can estimate breakdown locations and durations based on for any specified machine alarm. Resolving these problems can improve reliability of automated machines despite their age.
Data gathering, machine learning and analysis was done in cloud-based solution for better scalability, as well as to make analyzing of any available data set easier. All used data originates from operating container terminals. Power BI was used to create monthly breakdown reports for management and product line team.
Developed linear regression and hybrid of two different clustering methods are used to estimate common breakdown locations and downtime caused by alarms. Two different downtime estimating models are created: one for complete terminal area, and second for crucial backreach area. Clustering used to find common breakdown locations uses the combination of density based DBSCAN and common k-means methods. Further, extensive automated Straddle Carrier alarm analysis was performed over two month’s data.
Pareto Principle was used to identify major trouble causing alarms to maximize efficient product development for improved customer usability. Used models can be used with any sized data set for any environment to troubleshoot terminal specific issues in e.g. terminals Wi-Fi coverage. As such, it can provide valuable information for customers and product development teams to improve automated machines reliability. The carried out alarm analysis found multiple improvement points, which were reported to product development. Monthly alarm analysis is ongoing and first addressed items have vanished from alarm analysis thus leading to decreased number of alarms.
Data gathering, machine learning and analysis was done in cloud-based solution for better scalability, as well as to make analyzing of any available data set easier. All used data originates from operating container terminals. Power BI was used to create monthly breakdown reports for management and product line team.
Developed linear regression and hybrid of two different clustering methods are used to estimate common breakdown locations and downtime caused by alarms. Two different downtime estimating models are created: one for complete terminal area, and second for crucial backreach area. Clustering used to find common breakdown locations uses the combination of density based DBSCAN and common k-means methods. Further, extensive automated Straddle Carrier alarm analysis was performed over two month’s data.
Pareto Principle was used to identify major trouble causing alarms to maximize efficient product development for improved customer usability. Used models can be used with any sized data set for any environment to troubleshoot terminal specific issues in e.g. terminals Wi-Fi coverage. As such, it can provide valuable information for customers and product development teams to improve automated machines reliability. The carried out alarm analysis found multiple improvement points, which were reported to product development. Monthly alarm analysis is ongoing and first addressed items have vanished from alarm analysis thus leading to decreased number of alarms.