Publication detail

Predicting Safety Solutions via an Artificial Neural Network

ŠTOHL, R. STIBOR, K.

Original Title

Predicting Safety Solutions via an Artificial Neural Network

English Title

Predicting Safety Solutions via an Artificial Neural Network

Type

conference paper

Language

en

Original Abstract

Considering the extensive data sets and statistical techniques, Industry 4.0 embodies a branch of machine learning that has a constantly increasing impact on machine safety. We propose an preliminary study based on application of multi-layer feed-forward neural networks in machine safety solutions; the approach is expected to simplify the user choice of suitable measures and safety functions. The prediction method and factors influencing the success rate of the procedure are indicated in a safety parameter scale reflecting industrial experience with classic methods. The multilayer perceptron, a mainstream classification algorithm from the WEKA machine learning workbench, was employed in our primary dataset as a class of the feed-forward artificial neural network. Our initial experimental data were collected from various experts within the industry. The overall proportion of individual safety solutions was correctly assigned by using the training-evaluated test mode, and its prediction accuracy was 100%; further, when assessing the 5-fold cross-validation test mode, we obtained the success rate of 40%. These statistical tools could be used to assess safety PLC traceability systems, and they exhibit the potential to assist managers in decision-making as safety devices. We demonstrate that machine learning is widely usable by the expert community and might bring multiple advantages, such as reduction of the safety solution design time, major cost cutback, and engineering tool availability.

English abstract

Considering the extensive data sets and statistical techniques, Industry 4.0 embodies a branch of machine learning that has a constantly increasing impact on machine safety. We propose an preliminary study based on application of multi-layer feed-forward neural networks in machine safety solutions; the approach is expected to simplify the user choice of suitable measures and safety functions. The prediction method and factors influencing the success rate of the procedure are indicated in a safety parameter scale reflecting industrial experience with classic methods. The multilayer perceptron, a mainstream classification algorithm from the WEKA machine learning workbench, was employed in our primary dataset as a class of the feed-forward artificial neural network. Our initial experimental data were collected from various experts within the industry. The overall proportion of individual safety solutions was correctly assigned by using the training-evaluated test mode, and its prediction accuracy was 100%; further, when assessing the 5-fold cross-validation test mode, we obtained the success rate of 40%. These statistical tools could be used to assess safety PLC traceability systems, and they exhibit the potential to assist managers in decision-making as safety devices. We demonstrate that machine learning is widely usable by the expert community and might bring multiple advantages, such as reduction of the safety solution design time, major cost cutback, and engineering tool availability.

Keywords

Assignment success, safety, risk assessment, artificial neural network, Industry 4.0

Released

29.10.2019

Publisher

IFAC-PapersOnLine

Pages from

490

Pages to

495

Pages count

6

URL

Documents

BibTex


@inproceedings{BUT159872,
  author="Radek {Štohl} and Karel {Stibor}",
  title="Predicting Safety Solutions via an Artificial Neural Network",
  annote="Considering the extensive data sets and statistical techniques, Industry 4.0 embodies a branch of machine learning that has a constantly increasing impact on machine safety. We propose an preliminary study based on application of multi-layer feed-forward neural networks in machine safety solutions; the approach is expected to simplify the user choice of suitable measures and safety functions. The prediction method and factors influencing the success rate of the procedure are indicated in a safety parameter scale reflecting industrial experience with classic methods. The multilayer perceptron, a mainstream classification algorithm from the WEKA machine learning workbench, was employed in our primary dataset as a class of the feed-forward artificial neural network. Our initial experimental data were collected from various experts within the industry. The overall proportion of individual safety solutions was correctly assigned by using the training-evaluated test mode, and its prediction accuracy was 100%; further, when assessing the 5-fold cross-validation test mode, we obtained the success rate of 40%. These statistical tools could be used to assess safety PLC traceability systems, and they exhibit the potential to assist managers in decision-making as safety devices. We demonstrate that machine learning is widely usable by the expert community and might bring multiple advantages, such as reduction of the safety solution design time,  major cost cutback, and engineering tool availability.",
  address="IFAC-PapersOnLine",
  booktitle="16th IFAC Conference on Programmable Devices and Embedded Systems PDeS 2019",
  chapter="159872",
  doi="10.1016/j.ifacol.2019.12.711",
  howpublished="online",
  institution="IFAC-PapersOnLine",
  number="27",
  year="2019",
  month="october",
  pages="490--495",
  publisher="IFAC-PapersOnLine",
  type="conference paper"
}