Details
Original language | English |
---|---|
Title of host publication | 2021 IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM) |
Pages | 1297-1302 |
Number of pages | 6 |
ISBN (electronic) | 978-1-6654-4139-1 |
Publication status | Published - 2021 |
Publication series
Name | IEEE/ASME International Conference on Advanced Intelligent Mechatronics, AIM |
---|---|
Volume | 2021-July |
Abstract
As key components in the field of industry 4.0, data of sensors is often used for checking and observing the quality of subsystems. In modern manufacturing environments this huge amount of data enables machine health monitoring tools to analyze the behavior of mechanical components over time e.g. to estimate the remaining useful life (RUL) before breakdown. In this paper a system based on an autoencoder alike structure to forecast the deterioration of components is introduced. It is capable to predict the RUL based on the historical stress and usage conditions and identify anomalies like occurring faults by predicting the future with the encoder part, projecting it backwards with the decoder part, and then comparing it with the original data. The degradation forecast is estimated with respect to direct measurable parameters and not using a virtual health index. Our approach estimates the RUL on limited and noisy data and does not require knowledge of the true RUL. With the proposed setup our model is scalable to other production line configurations and product derivatives with different given production or quality thresholds without the need of a new training. We use real process data as well as synthetic signals for the training of the neural networks to improve the performance. We evaluate and demonstrate the performance of our RUL estimation approach against established forecast methods in the field of glass forming processes. We show that our approach of time series prediction in comparison to established prediction methods like RANSAC or ARIMA which require background knowledge delivers comparable accuracy and can additionally predict abnormal behavior.
ASJC Scopus subject areas
- Engineering(all)
- Electrical and Electronic Engineering
- Engineering(all)
- Control and Systems Engineering
- Computer Science(all)
- Computer Science Applications
- Computer Science(all)
- Software
Cite this
- Standard
- Harvard
- Apa
- Vancouver
- BibTeX
- RIS
2021 IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM). 2021. p. 1297-1302 (IEEE/ASME International Conference on Advanced Intelligent Mechatronics, AIM; Vol. 2021-July).
Research output: Chapter in book/report/conference proceeding › Conference contribution › Transfer › peer review
}
TY - GEN
T1 - Introduction to an Adaptive Remaining Useful Life Prediction for forming tools
AU - Kellermann, Christoph
AU - Munderloh, Marco
AU - Neumann, Eric
AU - Adhisantoso, Yeremia Gunawan
AU - Ostermann, Jörn
PY - 2021
Y1 - 2021
N2 - As key components in the field of industry 4.0, data of sensors is often used for checking and observing the quality of subsystems. In modern manufacturing environments this huge amount of data enables machine health monitoring tools to analyze the behavior of mechanical components over time e.g. to estimate the remaining useful life (RUL) before breakdown. In this paper a system based on an autoencoder alike structure to forecast the deterioration of components is introduced. It is capable to predict the RUL based on the historical stress and usage conditions and identify anomalies like occurring faults by predicting the future with the encoder part, projecting it backwards with the decoder part, and then comparing it with the original data. The degradation forecast is estimated with respect to direct measurable parameters and not using a virtual health index. Our approach estimates the RUL on limited and noisy data and does not require knowledge of the true RUL. With the proposed setup our model is scalable to other production line configurations and product derivatives with different given production or quality thresholds without the need of a new training. We use real process data as well as synthetic signals for the training of the neural networks to improve the performance. We evaluate and demonstrate the performance of our RUL estimation approach against established forecast methods in the field of glass forming processes. We show that our approach of time series prediction in comparison to established prediction methods like RANSAC or ARIMA which require background knowledge delivers comparable accuracy and can additionally predict abnormal behavior.
AB - As key components in the field of industry 4.0, data of sensors is often used for checking and observing the quality of subsystems. In modern manufacturing environments this huge amount of data enables machine health monitoring tools to analyze the behavior of mechanical components over time e.g. to estimate the remaining useful life (RUL) before breakdown. In this paper a system based on an autoencoder alike structure to forecast the deterioration of components is introduced. It is capable to predict the RUL based on the historical stress and usage conditions and identify anomalies like occurring faults by predicting the future with the encoder part, projecting it backwards with the decoder part, and then comparing it with the original data. The degradation forecast is estimated with respect to direct measurable parameters and not using a virtual health index. Our approach estimates the RUL on limited and noisy data and does not require knowledge of the true RUL. With the proposed setup our model is scalable to other production line configurations and product derivatives with different given production or quality thresholds without the need of a new training. We use real process data as well as synthetic signals for the training of the neural networks to improve the performance. We evaluate and demonstrate the performance of our RUL estimation approach against established forecast methods in the field of glass forming processes. We show that our approach of time series prediction in comparison to established prediction methods like RANSAC or ARIMA which require background knowledge delivers comparable accuracy and can additionally predict abnormal behavior.
UR - http://www.scopus.com/inward/record.url?scp=85114961508&partnerID=8YFLogxK
U2 - 10.1109/aim46487.2021.9517557
DO - 10.1109/aim46487.2021.9517557
M3 - Conference contribution
SN - 978-1-6654-4140-7
T3 - IEEE/ASME International Conference on Advanced Intelligent Mechatronics, AIM
SP - 1297
EP - 1302
BT - 2021 IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM)
ER -