Improvement of Aircraft System Availability at Flight Operator: A350 Hydraulic System Case Study
Roukala, Elina (2020)
2020
Konetekniikan DI-ohjelma - Master's Programme in Mechanical Engineering
Tekniikan ja luonnontieteiden tiedekunta - Faculty of Engineering and Natural Sciences
This publication is copyrighted. Only for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2020-10-26
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202009307203
https://urn.fi/URN:NBN:fi:tuni-202009307203
Tiivistelmä
The on-time performance of flights is extremely important in commercial aviation, as delays can have a significant financial impact. Some delay causes, such as weather conditions, can’t be controlled. Therefore there’s even more pressure to prevent the delays that can be avoided. The purpose of this study was to develop a systematic method to improve an aircraft system’s availability by preventing delays caused by a technical failure at departure. The method was tested as a case study on the hydraulic system of Airbus A350. The case study was conducted on historical data, as it was not possible to implement it to contemporary data within the timescale of this study.
A modern commercial aircraft produces a considerable amount of condition monitoring data during each flight, but it’s not yet utilized to its full potential. It can be a challenge to choose parameters for successful condition monitoring from a large amount of data. Handling and analyzing large data masses can also be demanding. This thesis started by comparing three qualitative analysis methods to define the best suitable one for analyzing the failures of aircraft systems. Based on this comparison, it was decided to move forward with fault tree analysis, which is a commonly used standardized method for analyzing complex systems.
The parameters included in condition monitoring were based on the fault indications identified with the qualitative analysis combined with literary research. For the data analysis method, a pattern recognizing artificial neural network (ANN) was chosen. The network was trained to identify different fault indications from flight data recordings. The advances of an ANN are the possibility for retraining in the future and at least equivalent performance with statistical methods with less effort. A pattern recognizing feed-forward type of network is simple and possible to build with a limited amount of data while still achieving adequate reliability. Aircraft are designed as extremely reliable products and therefore using a more complex method would require a long data gathering period to gain enough data from different faults.
Promising results were achieved with the case study system with this method of combining qualitative analysis with data analysis. The neural network taught with features extracted from 150 flight data recordings learned to recognize two hydraulic system’s fault types from the data with sufficient reliability. The classification was successful up to 22 days before the fault led to actual failure. If more data from different fault types were available, the network could be retrained to more versatile classification and it could identify faults even earlier. Naturally, mere identification of imminent faults from data is not enough to improve a system’s availability, but the neural network would be a useful tool for the maintenance organization. Had the imminent faults identified from the case study system by the neural network led to maintenance actions, it would’ve been possible to prevent up to 30 % of the disruption events caused by the said system during the study period.
A modern commercial aircraft produces a considerable amount of condition monitoring data during each flight, but it’s not yet utilized to its full potential. It can be a challenge to choose parameters for successful condition monitoring from a large amount of data. Handling and analyzing large data masses can also be demanding. This thesis started by comparing three qualitative analysis methods to define the best suitable one for analyzing the failures of aircraft systems. Based on this comparison, it was decided to move forward with fault tree analysis, which is a commonly used standardized method for analyzing complex systems.
The parameters included in condition monitoring were based on the fault indications identified with the qualitative analysis combined with literary research. For the data analysis method, a pattern recognizing artificial neural network (ANN) was chosen. The network was trained to identify different fault indications from flight data recordings. The advances of an ANN are the possibility for retraining in the future and at least equivalent performance with statistical methods with less effort. A pattern recognizing feed-forward type of network is simple and possible to build with a limited amount of data while still achieving adequate reliability. Aircraft are designed as extremely reliable products and therefore using a more complex method would require a long data gathering period to gain enough data from different faults.
Promising results were achieved with the case study system with this method of combining qualitative analysis with data analysis. The neural network taught with features extracted from 150 flight data recordings learned to recognize two hydraulic system’s fault types from the data with sufficient reliability. The classification was successful up to 22 days before the fault led to actual failure. If more data from different fault types were available, the network could be retrained to more versatile classification and it could identify faults even earlier. Naturally, mere identification of imminent faults from data is not enough to improve a system’s availability, but the neural network would be a useful tool for the maintenance organization. Had the imminent faults identified from the case study system by the neural network led to maintenance actions, it would’ve been possible to prevent up to 30 % of the disruption events caused by the said system during the study period.