Session: 14-03-01: Reliability and Safety in Industrial Automation Systems

Paper Number: 70258

Start Time: Monday, 05:50 PM

70258 - KrakenBox: Deep Learning-Based Error Detector for Industrial Cyber-Physical Systems 

Online error detection helps to reduce the risk of failure of safety-critical systems. However, due to the increasing complexity of modern Cyber-Physical Systems and the sophisticated interaction of their heterogeneous components, it becomes harder to apply traditional error detection methods. Nowadays, the popularity of Deep Learning-based error detection snowballs. DL-based methods achieved significant progress along with better results.

This paper introduces the KrakenBox, a deep learning-based error detector for industrial Cyber-Physical Systems (CPS). It provides conceptual and technical details of the KrakenBox (i) Hardware, (ii) Software, and a case study (iii) DemoSystem.

The KrakenBox Hardware is based on NVIDIA Jetson AGX Xavier, designed to empower the deep learning-based application and the extended alarm module. The alarm module is equipped with a red/green lamp and a beeper to signal error detection. The KrakenBox Software consists of several programs capable of collecting, processing, storing, and analyzing time-series data. There are three primary deep learning-based time-series data error detection approaches: classification, reconstruction, and prediction. We have chosen the prediction approach since it gives us the advantage of online detection and localization. The prediction models of the KrakenBox software are developed with Tensorflow based on the Long Short-Term Memory (LSTM) recurrent neural network. In comparison to state-of-the-art, we used a light-weighted LSTM to achieve real-time properties. We can process three signals in real-time with a sample time of 0.1 seconds. During training, the prediction models are trained with error-free data, while during testing, we use erroneous data as input to the prediction model. For that, we inject common CPS faults such as stuck-at, package drop, bias/offset, bit flips, time delay, and noise to the system, which make data erroneous. We raise the red flag if the residual of the actual and predicted values is higher than a certain threshold. In order to demonstrate the applicability and the performance of our error detector, we exploited a Simulink model of two industrial manipulators as the DemoSystem equipped with fault injectors. In order to generate training data, we monitor and log the signals of the position sensors through simulation. During the simulation, random waypoints are generated for the manipulators to emulate different kinds of manipulation. Model-based fault injection gave us the possibility to inject faults flexibly with deterministic and stochastic approaches. The injected faults with different types, magnitudes, and locations can be customized through a user-friendly UI.

A modern industrial automation system consists of different kinds of components networked with each other. The KrakenBox can be connected to the networked automation system either via Ethernet or wirelessly. The paper describes extensive experiments that allow the evaluation of the error detection performance of the KrakenBox for varying error magnitude. The results of these experiments demonstrate that the KrakenBox is able to significantly improve the safety of a networked automation system.

Presenting Author: Sheng Ding University of Stuttgart, Institute of Industrial Automation and Software Engineering

Authors:

Sheng Ding University of Stuttgart
Andrey Morozov University of Stuttgart
Tagir Fabarisov University of Stuttgart
Silvia Vock Bundesanstalt fur Arbeitsschutz und Arbeitsmedizin