Structured Denoising Autoencoder for Fault Detection and Analysis

Takaaki Tagawa, Yukihiro Tadokoro, Takehisa Yairi
Proceedings of the Sixth Asian Conference on Machine Learning, PMLR 39:96-111, 2015.

Abstract

This paper proposes a new fault detection and analysis approach which can leverage incomplete prior information. Conventional data-driven approaches suffer from the problem of overfitting and result in high rates of false positives, and model-driven approaches suffer from a lack of specific information about complex systems. We overcome these problems by modifying the denoising autoencoder (DA), a data-driven method, to form a new approach, called the structured denoising autoencoder (SDA), which can utilize incomplete prior information. The SDA does not require specific information and can perform well without overfitting. In particular, an empirical analysis with synthetic data revealed that the SDA performs better than the DA even when there is partially incorrect or abstract information. An evaluation using real data from moving cars also showed that the SDA with incomplete knowledge outperformed conventional methods. Surprisingly, the SDA results were better even though the parameters of the conventional methods were tuned using faulty data, which are normally unknown. In addition, the SDA fault analysis was able to extract the true causes of the changes within the faulty data; the other methods were unable to do this. Thus, only our proposed method can explain why the faults occurred.

Cite this Paper

BibTeX
@InProceedings{pmlr-v39-tagawa14, title = {Structured Denoising Autoencoder for Fault Detection and Analysis}, author = {Tagawa, Takaaki and Tadokoro, Yukihiro and Yairi, Takehisa}, booktitle = {Proceedings of the Sixth Asian Conference on Machine Learning}, pages = {96--111}, year = {2015}, editor = {Phung, Dinh and Li, Hang}, volume = {39}, series = {Proceedings of Machine Learning Research}, address = {Nha Trang City, Vietnam}, month = {26--28 Nov}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v39/tagawa14.pdf}, url = {https://proceedings.mlr.press/v39/tagawa14.html}, abstract = {This paper proposes a new fault detection and analysis approach which can leverage incomplete prior information. Conventional data-driven approaches suffer from the problem of overfitting and result in high rates of false positives, and model-driven approaches suffer from a lack of specific information about complex systems. We overcome these problems by modifying the denoising autoencoder (DA), a data-driven method, to form a new approach, called the structured denoising autoencoder (SDA), which can utilize incomplete prior information. The SDA does not require specific information and can perform well without overfitting. In particular, an empirical analysis with synthetic data revealed that the SDA performs better than the DA even when there is partially incorrect or abstract information. An evaluation using real data from moving cars also showed that the SDA with incomplete knowledge outperformed conventional methods. Surprisingly, the SDA results were better even though the parameters of the conventional methods were tuned using faulty data, which are normally unknown. In addition, the SDA fault analysis was able to extract the true causes of the changes within the faulty data; the other methods were unable to do this. Thus, only our proposed method can explain why the faults occurred.} }
Endnote
%0 Conference Paper %T Structured Denoising Autoencoder for Fault Detection and Analysis %A Takaaki Tagawa %A Yukihiro Tadokoro %A Takehisa Yairi %B Proceedings of the Sixth Asian Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2015 %E Dinh Phung %E Hang Li %F pmlr-v39-tagawa14 %I PMLR %P 96--111 %U https://proceedings.mlr.press/v39/tagawa14.html %V 39 %X This paper proposes a new fault detection and analysis approach which can leverage incomplete prior information. Conventional data-driven approaches suffer from the problem of overfitting and result in high rates of false positives, and model-driven approaches suffer from a lack of specific information about complex systems. We overcome these problems by modifying the denoising autoencoder (DA), a data-driven method, to form a new approach, called the structured denoising autoencoder (SDA), which can utilize incomplete prior information. The SDA does not require specific information and can perform well without overfitting. In particular, an empirical analysis with synthetic data revealed that the SDA performs better than the DA even when there is partially incorrect or abstract information. An evaluation using real data from moving cars also showed that the SDA with incomplete knowledge outperformed conventional methods. Surprisingly, the SDA results were better even though the parameters of the conventional methods were tuned using faulty data, which are normally unknown. In addition, the SDA fault analysis was able to extract the true causes of the changes within the faulty data; the other methods were unable to do this. Thus, only our proposed method can explain why the faults occurred.
RIS
TY - CPAPER TI - Structured Denoising Autoencoder for Fault Detection and Analysis AU - Takaaki Tagawa AU - Yukihiro Tadokoro AU - Takehisa Yairi BT - Proceedings of the Sixth Asian Conference on Machine Learning DA - 2015/02/16 ED - Dinh Phung ED - Hang Li ID - pmlr-v39-tagawa14 PB - PMLR DP - Proceedings of Machine Learning Research VL - 39 SP - 96 EP - 111 L1 - http://proceedings.mlr.press/v39/tagawa14.pdf UR - https://proceedings.mlr.press/v39/tagawa14.html AB - This paper proposes a new fault detection and analysis approach which can leverage incomplete prior information. Conventional data-driven approaches suffer from the problem of overfitting and result in high rates of false positives, and model-driven approaches suffer from a lack of specific information about complex systems. We overcome these problems by modifying the denoising autoencoder (DA), a data-driven method, to form a new approach, called the structured denoising autoencoder (SDA), which can utilize incomplete prior information. The SDA does not require specific information and can perform well without overfitting. In particular, an empirical analysis with synthetic data revealed that the SDA performs better than the DA even when there is partially incorrect or abstract information. An evaluation using real data from moving cars also showed that the SDA with incomplete knowledge outperformed conventional methods. Surprisingly, the SDA results were better even though the parameters of the conventional methods were tuned using faulty data, which are normally unknown. In addition, the SDA fault analysis was able to extract the true causes of the changes within the faulty data; the other methods were unable to do this. Thus, only our proposed method can explain why the faults occurred. ER -
APA
Tagawa, T., Tadokoro, Y. & Yairi, T.. (2015). Structured Denoising Autoencoder for Fault Detection and Analysis. Proceedings of the Sixth Asian Conference on Machine Learning, in Proceedings of Machine Learning Research 39:96-111 Available from https://proceedings.mlr.press/v39/tagawa14.html.

Related Material