Predicting harmful conditions with Hidden Markov Models
| Type of publication: | Mastersthesis |
| Citation: | |
| Year: | 2017 |
| Month: | May |
| School: | TU Darmstadt |
| Abstract: | Predictive Maintenance is a technique used to predict the condition of in-service equipment for adapting a maintenance schedule. Techlok is a project of DB Cargo, benefiting from Predictive Maintenance in order to increase the trains availability and cost reduction. The trains are equipped with sensors to produce continuous log file of diagnostic data. Based on these diagnostic data, a scenario is employed to create a predictor of failure trains. Hidden Markov Model (HMM) is a Machine Learning algorithm rest on Markov Model with hidden states. The model is applied in different fields such as speech recognition and gen techniques. In this work dealing with lots of diagnostic data, Hidden Markov Model is used in order to develop a failure predictor model. The results show that using HMM to predict the failure with respect of the incoming system data is possible. The model classifies with an accuracy of 96%. The classifier can predict well, but it is strongly dependent to the data stabilities. Although here a subset of the whole diagnostic codes is examined, the results encourage generalizing the model forward all features. More research and development is called for. |
| Keywords: | |
| Authors | |
|
Attachments
|
|
|
|
|
Topics
|
|
|
|
