Maintenance cost estimation for load handling equipment: Maintenance cost estimation for reachstackers using gradient boost regressor estimator
Valinejad, Niloufar (2020)
2020
Tietotekniikan DI-ohjelma - Master's Programme in Information Technology
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2020-09-29
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202009086926
https://urn.fi/URN:NBN:fi:tuni-202009086926
Tiivistelmä
Virtually all heavy-duty machines need maintenance, and the maintenance cost can be significant for providers of cargo and load handling solutions. Professionals in such industries must be able to forecast this cost accurately as minimizing this cost helps assure reasonable profits for companies and improving their economy.
The cost for maintenance varies for different machines based on the payloads they carry, working environment, operator skills, among others. That makes the cost difficult to predict since we usually do not have access to such sources of information. The current practice of using statistical regression methods cannot suitably capture the relationship between the repair cost of heavy equipment and its influencing factors.
In this thesis, the potential of Machine Learning (ML) models was evaluated as an alternative method for the prediction of maintenance cost of load handling machines. The distinctive difference is discovering the possibility of predicting this cost based on telemetry data merged with machine details, also analyzing parameters affects this cost the most.
This study was conducted based on data received from 483 Kalmar’s reachstacker’s since 2014 during their service contract or warranty contract. First, a detailed analysis of the historical data allows identifying the distributions of maintenance expenses and their fluctuated patterns during different RS’ life periods. Then the research continued by the implementation of a treebased ML model to predict two different predictive variables; 1) Cumulative maintenance cost per engine working hour (CMCPH) and 2) Cumulative maintenance cost per lift (CMCPL).
The results of the ML approach show better interpretability and adequate accuracy by considering CMCPL as the output variable with Meter per lift, fuel used per lift, and tons per lift as the most influential predictors of Maintenance Cost. One surprising observation was having the length of the service work order as one of the topmost important features affecting the result of the experiment. An accurate prediction of future equipment maintenance costs can promote decision-making tasks related to equipment budget and resource planning by injecting more observations to the model to decrease the variance.
The cost for maintenance varies for different machines based on the payloads they carry, working environment, operator skills, among others. That makes the cost difficult to predict since we usually do not have access to such sources of information. The current practice of using statistical regression methods cannot suitably capture the relationship between the repair cost of heavy equipment and its influencing factors.
In this thesis, the potential of Machine Learning (ML) models was evaluated as an alternative method for the prediction of maintenance cost of load handling machines. The distinctive difference is discovering the possibility of predicting this cost based on telemetry data merged with machine details, also analyzing parameters affects this cost the most.
This study was conducted based on data received from 483 Kalmar’s reachstacker’s since 2014 during their service contract or warranty contract. First, a detailed analysis of the historical data allows identifying the distributions of maintenance expenses and their fluctuated patterns during different RS’ life periods. Then the research continued by the implementation of a treebased ML model to predict two different predictive variables; 1) Cumulative maintenance cost per engine working hour (CMCPH) and 2) Cumulative maintenance cost per lift (CMCPL).
The results of the ML approach show better interpretability and adequate accuracy by considering CMCPL as the output variable with Meter per lift, fuel used per lift, and tons per lift as the most influential predictors of Maintenance Cost. One surprising observation was having the length of the service work order as one of the topmost important features affecting the result of the experiment. An accurate prediction of future equipment maintenance costs can promote decision-making tasks related to equipment budget and resource planning by injecting more observations to the model to decrease the variance.