Technical Stream
Technical Stream
Intelligent process monitoring and online fault diagnosis for prototype cyber-physical system (CPS)


In the context of elementary cyber-physical system the predictive model of correct system functioning should be developed and system faults should be recognized using process monitoring data. The proposed approach should be potentially applicable to more complex and realistic industrial CPS.

1. Statement of the problem.

Consider prototype CPS that represents the reduced replica of elementary drill conveyor system illustrated in Fig. 1. The functioning of this CPS is performed by the following scenario: the wooden workpiece (W) in the initial position is dislocated at the trolley-protractor (TP) which, in turn, at the beginning of technological cycle is positioned in the right-bottom corner of the prototype. Activating of the CPS leads to moving of TP with the workpiece W from right to the leftmost position, then transferring of W from the TP to the left protractor (LP), moving of W on LP to the left rotating protractor (LRP), clockwise rotating of LRP by 90 degrees and moving to drill mechanism (DM). After imitation of drilling procedure of W at DM its protractor forwards W to the second (right) rotating protractor (RRP), then RRP clockwise rotates by 90 degrees and transfers W to the right protractor (RP) which is finally conveys W to TP under condition of its returning into initial position. Movable protractors TP, LRP and RRP return into their initial positions as soon as the workpiece W leaves the corresponding equipment.

Intelligence process monitoring and online fault diagnosis for prototype cyber-physical system (CPS)

Fig.1.The illustration of drill conveyer prototype. The workpiece W and the protractors TP, LRP and RRP are represented in their initial positions.

The operation control of this CPS is performed by the set of controllers which realize aforementioned conveyor logic using, in particular, data from the sensors (see Fig.1). Along with the correct CPS behavior occasionally in the system functioning process can be occurred abnormal situations having natural reasons (for example, poor fitting of equipment parts) as well as caused by intentional attempts of CPS logic cracking with the aim of equipment’s damaging. The objective of current investigation is to develop online intelligent monitoring system capable of preventing abnormal situations in CPS using solely control and sensor time series.

2. Data description.

In process of CPS operation all control and sensor data are stored at the system hard drive and can be available for the analysis including online mode. The researchers are provided by approximately 20 control and 20 sensor time series without specifying which series belongs to which data type. Total observation duration is T ~ 1 hour and the duration of one CPS technological cycle is Tc ~ 30 seconds. It is assumed that data at this time interval does not contain any system faults and represents variation in correct CPS behavior. The data was written in frequently-used “csv” format so that each “csv”-file contains timestamps vs values corresponding to some time series representing control or sensor variables.

3. Requirements for CPS monitoring system.

Safe functioning of CPS is supposed to be implemented based on intelligent predictive model for the aforementioned control/sensor time series. This model has to be able to predict system behavior at some time ahead, recognize abnormal CPS behavior and  interpret it in terms of deviations of the current time series evolution from the the forecast. The testing of the model can be performed using two scenarios: in absence of CPS faults the model’s forecast should provide adequate description of time series behavior at least at time 2Tc  and in the presence of these faults should ensure their robust detection.

Under conditions of successfully testing of the model at the historical data set (see sect. 2) as the additional results researchers should provide:

  1. The forecast of historical time series at the duration at least time 2Tc starting from the maximal timestamp from all series;
  2. Complete description of the proposed model (Word, PowerPoint, LaTeX, etc.);
  3. The model implementation and algorithms (pseudo-code, Python, MatLab, etc.).

The testing of the model in fault detection mode will be performed at the additional competition stage after evaluating  provided solution on items 1-3.

4. Evaluation criteria.

The models evaluation will be performed using following basic criteria:

  1. Prediction quality of the model in fault-free regime of CPS functioning;
  2. Suitability and novelty of proposed approach, the scalability of proposed approach to real industrial CPS;
  3. Robust CPS fault detection.

5. Supplementary materials (available after registration)

  1. Historical data set for drill conveyer prototype
  2. The movie of drill conveyer prototype operation
  3. The snapshot of drill conveyer prototype
Interested? Ask an expert.
Register and start now!
Anti-Malware Research