liu.seSearch for publications in DiVA
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • oxford
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Variable selection for heavy-duty vehicle battery failure prognostics using random survival forests
Linköping University, Department of Electrical Engineering, Vehicular Systems. Linköping University, Faculty of Science & Engineering.
Linköping University, Department of Electrical Engineering, Vehicular Systems. Linköping University, Faculty of Science & Engineering.
Linköping University, Department of Electrical Engineering, Vehicular Systems. Linköping University, Faculty of Science & Engineering.
2016 (English)In: PHME 2016 Proceedings of the Third European Conference of the Prognostics and Health Management Society 2016, Bilbao, Spain July 5–8, 2016 / [ed] Ioana Eballard and Anibal Bregon, 2016, p. 649-659Conference paper, Published paper (Refereed)
Abstract [en]

Prognostics and health management is a useful tool for more flexible maintenance planning and increased system reliability. The application in this study is lead-acid battery failure prognosis for heavy-duty trucks which is important to avoid unplanned stops by the road. There are large amounts of data available, logged from trucks in operation. However, datais not closely related to battery health which makes battery prognostic challenging. When developing a data-driven prognostics model and the number of available variables is large,variable selection is an important task, since including non-informative variables in the model have a negative impact on prognosis performance. Two features of the dataset has been identified, 1) few informative variables, and 2) highly correlated variables in the dataset. The main contribution is a novel method for identifying important variables, taking these two properties into account, using Random Survival Forests to estimate prognostics models. The result of the proposed method is compared to existing variable selection methods,and applied to a real-world automotive dataset. Prognostic models with all and reduced set of variables are generated and differences between the model predictions are discussed, and favorable properties of the proposed approach are highlighted.

Place, publisher, year, edition, pages
2016. p. 649-659
Keywords [en]
variable selection, random survival forest, battery failure prognostics
National Category
Other Electrical Engineering, Electronic Engineering, Information Engineering
Identifiers
URN: urn:nbn:se:liu:diva-131788ISBN: 978-1-936263-21-9 (print)OAI: oai:DiVA.org:liu-131788DiVA, id: diva2:1033379
Conference
Third European Conference of the Prognostics and Health Management Society 2016, Bilbao, July 5-8, 2016
Available from: 2016-10-06 Created: 2016-10-06 Last updated: 2020-01-24Bibliographically approved
In thesis
1. Data-driven lead-acid battery lifetime prognostics
Open this publication in new window or tab >>Data-driven lead-acid battery lifetime prognostics
2017 (English)Licentiate thesis, comprehensive summary (Other academic)
Abstract [en]

To efficiently transport goods by heavy-duty trucks, it is important that vehicles have a high degree of availability and in particular avoid becoming standing by the road unable to continue the transport mission. An unplanned stop by the road does not only cost due to the delay in delivery, but can also lead to a damaged cargo. High availability can be achieved by changing components frequently, but such an approach is expensive both due to the frequent visits to a workshop and also due to the component cost. Therefore, failure prognostics and flexible maintenance has significant potential in the automotive field for both manufacturers, commercial fleet owners, and private customers.

In heavy-duty trucks, one cause of unplanned stops are failures in the electrical power system, and in particular the lead-acid starter battery. The main purpose of the battery is to power the starter motor to get the diesel engine running, but it is also used to, for example, power auxiliary units such as cabin heating and kitchen equipment. Detailed physical models of battery degradation is inherently difficult and requires, in addition to battery health sensing which is not available in the given study, detailed knowledge of battery chemistry and how degradation depends on the vehicle and battery usage profiles.

The main aim of the given work is to predict the lifetime of lead-acid batteries using data-driven approaches. Main contributions in the thesis are: a) the choice of the Random Survival Forest method as the model for predicting a conditional reliability function which is used as the estimator of the battery lifetime, b) variable selection for better predictability of the model and c) variance estimation for the Random Survival Forest method.

When developing a data-driven prognostic model and the number of available variables is large, variable selection is an important task, since including non-informative variables in the model have a negative impact on prognosis performance. Two features of the dataset has been identified, 1) there are few informative variables, and 2) highly correlated variables in the dataset. The main contribution is a novel method for identifying important variables, taking these two properties into account, using Random Survival Forests to estimate prognostics models. The result of the proposed method is compared to existing variable selection methods, and applied to a real-world automotive dataset.

Confidence bands are introduced to the RSF model giving an opportunity for an engineer to observe the confidence of the model prediction. Some aspects of the confidence bands are considered: a) their asymptotic behavior and b) usefulness in the model selection. A problem of including time related variables is addressed in the thesis with arguments why it is a good choice not to add them into the model. Metrics for performance evaluation are suggested which show that the model can be used to find and optimize cost of the battery replacement.

Place, publisher, year, edition, pages
Linköping: Linköping University Electronic Press, 2017. p. 28
Series
Linköping Studies in Science and Technology. Thesis, ISSN 0280-7971 ; 1779
National Category
Probability Theory and Statistics Control Engineering Transport Systems and Logistics Economics and Business
Identifiers
urn:nbn:se:liu:diva-137526 (URN)978-91-7685-504-1 (ISBN)
Presentation
2017-05-31, Planck, D-huset, Campus Valla, Linköping, 10:15 (English)
Opponent
Supervisors
Available from: 2017-05-19 Created: 2017-05-19 Last updated: 2019-09-23Bibliographically approved
2. Machine Learning Models for Predictive Maintenance
Open this publication in new window or tab >>Machine Learning Models for Predictive Maintenance
2020 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

The amount of goods produced and transported around the world each year increases and heavy-duty trucks are an important link in the logistic chain. To guarantee reliable delivery a high degree of availability is required, i.e., avoid standing by the road unable to continue the transport mission. Unplanned stops by the road do not only cost due to the delay in delivery, but can also lead to damaged cargo. Vehicle downtime can be reduced by replacing components based on statistics of previous failures. However, such an approach is both expensive due to the required frequent visits to a workshop and inefficient as many components from the vehicles in the fleet are still operational. A prognostic method, allowing for vehicle individualized maintenance plans, therefore poses a significant potential in the automotive field. The prognostic method estimates component degradation and remaining useful life based on recorded data and how the vehicle has been operated.

Lead-acid batteries is a part of the electrical power system in a heavy-duty truck, primarily responsible for powering the starter motor but also powering auxiliary units, e.g., cabin heating and kitchen equipment, which makes the battery a vital component for vehicle availability. Developing physical models of battery degradation is a difficult process which requires access to battery health sensing that is not available in the given study as well a detailed knowledge of battery chemistry.

An alternative approach, considered in this work, is data-driven methods based on large amounts of logged data describing vehicle operation conditions. In the use-case studied, recorded data is not closely related to battery health which makes battery prognostic challenging. Data is collected during infrequent and non-equidistant visits to a workshop and there are complex dependencies between variables in the data. The main aim of this work has been to develop a framework and methods for estimating lifetime of lead-acid batteries using data-driven methods for condition-based maintenance. The methodology is general and can be applicable for prognostics of other components.

A main contribution of the thesis is development of machine learning models for predictive maintenance, estimating conditional reliability functions, using Random Survival Forests (RSF) and recurrent neural networks (RNN). An important property of the data is that for a specific vehicle there may be multiple data readouts, but also one single data readout which makes predictive modeling challenging and dealing with this situation is discussed for both RSF and neural networks models. Data quality is important when building data-driven models, and here the data is imbalanced since there are few battery failures relative to the number of vehicles. Further, the data includes many uninformative variables and among those that are informative, there are complex dependencies and correlation. Methods for selecting which data features to use in the model in this situation is also a key contribution. When a point estimation of the conditional reliability functions is available, it is of interest to know how uncertain the estimate is as it allows to take quality of the prediction into account when deciding on maintenance actions. A theory for estimating the variance of the RSF predictor is another contribution in the thesis. To conclude, the results show that Long Short-Term Memory networks, which is a type of RNN, is the most suitable for the vehicle operational data and give the best performance among methods evaluated in the thesis.

Abstract [sv]

Mängden gods som produceras och transporteras världen runt ökar och tunga fordon är en viktig del i logistikkedjan. För att garantera pålitliga leveranser krävs hög tillgänglighet hos fordonen genom att bland annat undvika oplanerade stopp längs vägen. Tid då fordonet ej är tillgängligt kan reduceras genom att byta ut komponenter baserat på statistik från tidigare fel. En sådan ansats kan dock vara dyr på grund av för täta besök på verkstäder samt att många komponenter fungerar avsevärt längre beroende på hur hårt komponenten använts. En prognostikmetod för individualiserade underhållsplaner har därför en stor potential i fordonsfältet. Prognostikmetoden uppskattar komponenters degradation och tillgänglig livstid baserat på registrerade data och hur fordonet har använts.

Blysyrabatterier är en del av det elektriska kraftsystemet i en lastbil, primärt ansvariga för att kraftsätta startmotor, men också för att ge kraft åt hjälpsystem som kabinvärme och köksutrustning, vilket betyder att batteriet är en viktig komponent för fordonets tillgänglighet. Att utveckla fysikaliska modeller för batteridegradation är svårt och kräver tillgång till mätdata direkt kopplat till batteriets hälsa, något som inte är tillgängligt i det här arbetet. En alternativ ansats, som utforskas här, är datadrivna metoder baserade på stora mängder inspelade data som beskriver hur fordonet använts. I studien är insamlad data ej direkt relaterad till batterihälsa vilket gör prognostikproblemet utmanande.

Ett huvudbidrag är utveckling av maskininlärningsmodeller för prediktivt underhåll baserad på Random Survival Forests (RSF) och Recurrent Neural Networks (RNN). En viktig egenskap hos insamlade data är att för specifika fordon så kan det finnas flera, eller endast enstaka, datautläsningar vilket också gör prediktiv modellering svårt. Metoder för att hantera detta för modeller baserade på RSF och neuronnät behandlas. Datakvalitet är viktigt vid utveckling av datadrivna modeller. Insamlade data är obalanserade eftersom det är få batterier som felar i relation till antalet fordon. Vidare, insamlade data inkluderar många oinformativa variabler och bland de informativa så finns komplexa beroenden och korrelationer. Metoder för att välja väl valda variabler att bygga modeller på för den här situationen är utmanande och ett huvudbidrag i arbetet. En central fråga är hur säker en punktskattning är och hur den osäkerheten kan vägas in när prediktiva underhållsplaner bestäms, speciellt när modellen baseras på så osäkra data och så ostrukturerade modeller som här. Ett viktigt bidrag är metodik för att estimera prediktionsvarians för RSF-modeller. Slutligen, ett huvudresultat för användarfallet är att LSTM-nät, ett typ av RNN, är den modellstruktur som ger bäst prestanda för prognostik av blysyrabatterier med det data som använts i avhandlingen.

Place, publisher, year, edition, pages
Linköping: Linköping University Electronic Press, 2020. p. 297
Series
Linköping Studies in Science and Technology. Dissertations, ISSN 0345-7524 ; 2040
National Category
Transport Systems and Logistics Computer Sciences Vehicle Engineering
Identifiers
urn:nbn:se:liu:diva-162649 (URN)10.3384/diss.diva-162649 (DOI)9789179299231 (ISBN)
Public defence
2020-03-06, Ada Lovelace, Building B, Campus US, Linköping, 10:15 (English)
Opponent
Supervisors
Available from: 2019-12-12 Created: 2019-12-12 Last updated: 2020-01-31Bibliographically approved

Open Access in DiVA

fulltext(649 kB)100 downloads
File information
File name FULLTEXT01.pdfFile size 649 kBChecksum SHA-512
84252b5f1a0de1337755385721d849c04b99ee285ee298f09be4bf26c13f568ba8c9bf30e38d6dbea54544fab282621d249cf44f4daa5de83ff9ca723adb9144
Type fulltextMimetype application/pdf

Search in DiVA

By author/editor
Voronov, SergiiJung, DanielFrisk, Erik
By organisation
Vehicular SystemsFaculty of Science & Engineering
Other Electrical Engineering, Electronic Engineering, Information Engineering

Search outside of DiVA

GoogleGoogle Scholar
Total: 100 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

isbn
urn-nbn

Altmetric score

isbn
urn-nbn
Total: 545 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • oxford
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf