Lifetime Predictions of Electrolytic Capacitors in Network Cameras with Random Forest
(2020)Centre for Mathematical Sciences
Department of Automatic Control
- Abstract
- Electrolytic capacitor components degrade when exposed to thermal stress which can cause failures in electrical devices. Several recent works have studied the lifetime of these components by using accelerated life testing. This work, however, takes a new approach by utilising vast amounts of temperature data and regression techniques. Axis Communications collects large amounts of non-personalised data from network cameras in real-time, which could be used for lifetime predictions.
However, there are various problems with the collected data, such as jitter, interruptions, and missing data. Methods to resolve these problems are developed and validated.
To predict the lifetime of an individual two different models are developed, a Random... (More) - Electrolytic capacitor components degrade when exposed to thermal stress which can cause failures in electrical devices. Several recent works have studied the lifetime of these components by using accelerated life testing. This work, however, takes a new approach by utilising vast amounts of temperature data and regression techniques. Axis Communications collects large amounts of non-personalised data from network cameras in real-time, which could be used for lifetime predictions.
However, there are various problems with the collected data, such as jitter, interruptions, and missing data. Methods to resolve these problems are developed and validated.
To predict the lifetime of an individual two different models are developed, a Random forest model and a Baseline model. The Baseline model is used as a validation of the performance of the Random forest model. The models require temperature data to create predictions. The goal is to achieve a mean absolute normalised error of less than 10 %, while simultaneously minimising the required amount of data. The Random forest model achieves the target mean absolute normalised error with
significantly less data than the Baseline model.
Furthermore, distributions of the lifetime predictions are formed, as they could help guide future product design. The distribution of the predictions is compared to the true distribution with statistical tests. The distribution of the Random forest predictions is concluded to be more similar to the true distribution than the Baseline model. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9029313
- author
- Andersson, Oscar and Hindgren, Oskar
- supervisor
- organization
- year
- 2020
- type
- H3 - Professional qualifications (4 Years - )
- subject
- report number
- TFRT-6114
- other publication id
- 0280-5316
- language
- English
- additional info
- FRTM01 Examensarbete i Reglerteknik
FMAM05 Examensarbete i matematik - id
- 9029313
- date added to LUP
- 2020-09-15 14:04:55
- date last changed
- 2020-09-15 14:39:49
@misc{9029313,
abstract = {{Electrolytic capacitor components degrade when exposed to thermal stress which can cause failures in electrical devices. Several recent works have studied the lifetime of these components by using accelerated life testing. This work, however, takes a new approach by utilising vast amounts of temperature data and regression techniques. Axis Communications collects large amounts of non-personalised data from network cameras in real-time, which could be used for lifetime predictions.
However, there are various problems with the collected data, such as jitter, interruptions, and missing data. Methods to resolve these problems are developed and validated.
To predict the lifetime of an individual two different models are developed, a Random forest model and a Baseline model. The Baseline model is used as a validation of the performance of the Random forest model. The models require temperature data to create predictions. The goal is to achieve a mean absolute normalised error of less than 10 %, while simultaneously minimising the required amount of data. The Random forest model achieves the target mean absolute normalised error with
significantly less data than the Baseline model.
Furthermore, distributions of the lifetime predictions are formed, as they could help guide future product design. The distribution of the predictions is compared to the true distribution with statistical tests. The distribution of the Random forest predictions is concluded to be more similar to the true distribution than the Baseline model.}},
author = {{Andersson, Oscar and Hindgren, Oskar}},
language = {{eng}},
note = {{Student Paper}},
title = {{Lifetime Predictions of Electrolytic Capacitors in Network Cameras with Random Forest}},
year = {{2020}},
}