Constructing robust health indicators from complex engineered systems via anticausal learning

Koutroulis, Georgios; Mutlu, Belgin; Kern, Roman

doi:10.1016/j.engappai.2022.104926

Cited by 21 publications

(12 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The article [9] was published this year and stated that there was a reduction of the average RMSE in all the investigated units by almost 65%, however, the results presented by the authors were inferior to this work. Note that the results presented for LSTM-AE (network resulting from the combination of Long short-term memory and auto-encoders), auto-encoders (AE), RR+ (least- squares with l2-regularization), and MLP+ (simple MLP architecture with l2-regularization) [9] in Table 6 are the average values of the 11, 14 and 15 results obtained from the paper. As seen, the CNN average results scored 4.86 with RMSE loss of 9.11, which was better than the winner [12], the second [3] and third [25] places of the 2021 PHM Competition [21].…”

Section: Resultscontrasting

confidence: 54%

“…In [9] a virtual health indicator was developed for degradation monitoring of safety-critical engineered systems that operate under time-varying conditions. It incorporates limited domain knowledge with a two-phase heuristics approach to select a causal driver and a set of measuring parameters.…”

Section: Related Workmentioning

confidence: 99%

“…The performance of all health indicators was tested for each of the units in terms of root mean square and the PHM of the RUL in cycles, it was not presented the average results. In [9] a framework with anti-causal learning is said to outperform existing deep learning architectures, reducing the average RMSE across all investigated units by nearly 65%.…”

Section: Related Workmentioning

confidence: 99%

“…This neural network has several hyper-parameters that can be adjusted, in addition to the sliding window size adjustment. The hyper-parameters of this work were defined based on what was found in the literature [3,25,9], but mainly by trial and error. There were 100 training epochs with batch size of 63 and an input batch format of [20,162,1].…”

Section: Convolutional Neural Network (Cnn)mentioning

confidence: 99%

See 3 more Smart Citations

Learning methods for remaining useful life prediction in a turbofan engine

Leal

Berton

Santos

2022

Anais Do XIX Encontro Nacional De Inteligência Artificial E Computacional (ENIAC 2022)

View full text Add to dashboard Cite

In industry 4.0 there is a growth in the Industrial Internet of Things (iIoT) with a lot of information generation and consequent big data challenges. Thus, it is imperative to have techniques able to process all this data and predict the maintenance of equipment and systems. The development of algorithms for remaining useful life (RUL) estimators is critical for the full functioning of the company’s assets. Especially the aeronautical sector needs to guarantee safety and quality flights. The turbofan, a propulsion engine, is a critical element for an airplane operation. This paper proposes a model to perform prediction of the remaining useful life of an aircraft’s turbo engine. In this work, we focus on the run-to-failure data from an N-CMAPSS turbofan, the data used were provided by NASA in 2021. After training and validating different algorithms such as MLP and CNN, we find CNN as the best approach with an RMSE of 9.11, a score of 5.14, and computed score of 1.17. The results have improved when compared to the literature over 25% in RMSE and 15% in computed score.

show abstract

Section: Resultscontrasting

confidence: 54%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Convolutional Neural Network (Cnn)mentioning

confidence: 99%

See 2 more Smart Citations

Learning methods for remaining useful life prediction in a turbofan engine

Leal

Berton

Santos

2022

Anais Do XIX Encontro Nacional De Inteligência Artificial E Computacional (ENIAC 2022)

View full text Add to dashboard Cite

show abstract

“…In general, a major drawback of such deep learning architectures is that they solely learn from correlations in the data. Therefore, causal structures that can render the model more robust under distribution shifts and noisy environments are likely to be overlooked [ 10 , 11 , 12 ]. Another inherent issue is the lack of interpretability of black-box models [ 13 ] since it is infeasible to understand the prediction mechanism for specific outputs.…”

Section: Introductionmentioning

confidence: 99%

A Causality-Inspired Approach for Anomaly Detection in a Water Treatment Testbed

Koutroulis

Mutlu

Kern

2022

Sensors

Self Cite

View full text Add to dashboard Cite

Critical infrastructure, such as water treatment facilities, largely relies on the effective functioning of industrial control systems (ICSs). Due to the wide adoption of high-speed network and digital infrastructure technologies, these systems are now highly interconnected not only to corporate networks but also to the public Internet, mostly for remote control and monitoring purposes. Sophisticated cyber-attacks may take advantage the increased interconnectedness or other security gaps of an ICS and infiltrate the system with devastating consequences to the economy, national security, and even human life. Due to the paramount importance of detecting and isolating these attacks, we propose an unsupervised anomaly detection approach that employs causal inference to construct a robust anomaly score in two phases. First, minimal domain knowledge via causal models helps identify critical interdependencies in the system, while univariate models contribute to individually learn the normal behavior of the system’s components. In the final phase, we employ the extreme studentized deviate (ESD) on the computed score to detect attacks and to exclude any irrelevant sensor signals. Our approach is validated on the widely used Secure Water Treatment (SWaT) benchmark, and it exhibits the highest F1 score with zero false alarms, which is extremely important for real-world deployment.

show abstract