Remaining Useful Life Prediction for Turbofan based on a Multilayer Perceptron and Kalman Filter

Alberto-Olivares, Mario; Gonzalez-Gutierrez, Arturo; Tovar-Arriaga, Saúl; Gorrostieta-Hurtado, Efrén

doi:10.1109/iceee.2019.8884495

Cited by 5 publications

(11 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…According to (Leto, 2008;Olivares, Gonzalez, Tovar, & Gorrostieta, 2019), there is no mathematical dependency between sensor data, which could support machine learning models training. Thus, (Leto, 2008) has proposed a normalization based on clustering of these six operating regimes, normalization which is refereed in Section 2.2.2.…”

Section: Pre-processing Datamentioning

confidence: 99%

“…The second degradation approach that illustrates a family of piece-wise curves, was adopted by (Heimes, 2008;Olivares et al, 2019) to reproduce the third winner technique (Leto, 2008) of the PHM08 Challenge (Ramasso & Saxena, 2014). This piece-wise approach corresponds to Eq.…”

Section: Rul Assumptionmentioning

confidence: 99%

“…Here, Rc is the initial constant value of RU L which varies from 120 to 130 cycles (Heimes, 2008); SoF is the time when engines start to failure is equal to t EoL − t i . After experimentation (Heimes, 2008;Olivares et al, 2019) limited Rc to take a value of 130, value which we adopted for our experiments.…”

Section: Rul Assumptionmentioning

confidence: 99%

“…Before the normalization step, a clustering process was carried out using K-means algorithm. This algorithm was used by (Ramasso & Saxena, 2014;Olivares et al, 2019) among others to add a regime identifier r for each operational cycle x. After that, a data normalization per regime N (.)…”

Section: Normalizationmentioning

confidence: 99%

“…For the sensor f on regime r, x (r,f ) represents the sensor data per regime, µ (r,f ) and σ (r,f ) are the mean and standard deviation, respectively. After that process, (Olivares et al, 2019) considered the normalization of the three operating settings, the cycle number and the target (RUL), without considering the regime variation. Given the performance improvements, we also apply the method in (Olivares et al, 2019).…”

Section: Normalizationmentioning

confidence: 99%

See 4 more Smart Citations

Remaining Useful Life Estimation in Aircraft Components with Federated Learning

2020

View full text Add to dashboard Cite

In this work, we compare distributed collaborative learning techniques for Prognostic and Health Management (PHM) systems, focusing on predictive aircraft maintenance. Aircraft industry components are usually evaluated using remaining useful life (RUL) estimations to describe the amount of time left before system health falls. Such estimates have been commonly achieved with traditional degradation estimation methods. These estimation methods have been widely applied in centralized processing architectures, limiting the scalability of PHM systems. Concerns about data privacy and transfer of large amount of data have also been limiting the construction of decentralized processing architectures. Nevertheless, with the emergence of collaborative training methods of machine learning models, e.g. Federated Learning (FL), the previous referred concerns have being tackled by privacy-preserving communications while keeping data at the network edges. However, the effectiveness of federated learning algorithms using time-series data for prognostic and health management of aircraft systems has been minimally explored. In this work, we use feed-forward neural networks on centralized and decentralized scenarios to compare the prediction error minimization of FL algorithms, such as, Federated Average (FedAvg) and Federated Proximal Term (FedProx). Our experiments take into account gradient descendent minimization and averaging weights of neural networks. Using FedAvg, we obtained similar prediction errors to the centralized scenario but presenting uncertain predictions along the aggregation iterations. On the other hand, using FedProx, the prediction error curve progressively decreases along the aggregation iterations if μ takes values ~ 0.01.

show abstract

Section: Pre-processing Datamentioning

confidence: 99%