Volume: 2, 2023
2nd International PhD Student’s Conference at the University of Life Sciences in Lublin, Poland:
ENVIRONMENT – PLANT – ANIMAL – PRODUCT
Abstract number: T015
DOI: https://doi.org/10.24326/ICDSUPL2.T015
Published online: 19 April 2023
ICDSUPL, 2, T015 (2023)
Lift cabin door drive failure detection using random forest algorithm
Jakub Gęca1*
1 Department of Computer Science, Faculty of Electrical Engineering and Computer Science, Lublin University of Technology, Nadbystrzycka 38D, 20-618, Lublin, Poland
* Corresponding author: j.geca@pollub.pl
Abstract
Failures in passenger lift systems can cause very negative consequences, not only financially, but also related to the safety of goods and personnel. Therefore, the ability to detect faults and identify the faulty component is becoming an extremely useful task from the machine manufacturer’s perspective. The use of machine learning algorithms for fault detection is a state-of-the-art solution that allows to identify which component has failed and then to implement condition-based maintenance. The decision tree model assumes that classification decisions are made based on answering successive questions about the features of a dataset. This allows the algorithm to develop a set of universal rules that lead to correct decision-making. The disadvantage of decision trees is that when there are outliers in the data (those that are impossible or very difficult to classify correctly because it contains contradicting features), the tree grows to a huge size trying to determine a rule for classifying the outlier samples. To overcome the weaknesses of a single decision tree, the random forest algorithm was developed, the idea of which is to combine multiple decision trees into a single ensemble classifier. This results in less sensitivity to under- and overfitting. Multiple trees (forest) are created at random, and the final classification is based on majority voting or by averaging probabilistic prediction. This paper presents the detection of faults in the cabin door drive of a passenger lift using the random forest algorithm. For this purpose, a test bench with a prototype of the cabin door was built. Then a database was created by measuring the drive parameters for normal operation and various types of faults. The acquired data was processed in such a way as to aggregate individual duty cycles into a single data record, after which the data was standardized. In addition, hyperparameter tuning of the random forest algorithm was carried out using cross-validation. The results show that the random forest algorithm achieves high performance in elevator cabin drive fault detection even despite the class imbalance in the dataset.
How to cite
J. Gęca, 2023. Lift cabin door drive failure detection using random forest algorithm. In: 2nd International PhD Student’s Conference at the University of Life Sciences in Lublin, Poland: Environment – Plant – Animal – Product. https://doi.org/10.24326/ICDSUPL2/.T015