Model-free condition monitoring with confidence

ABSTRACT

Computational Self-awareness can improve performance, robustness, and adaptivity of a system. As a key element of self-awareness, observation quality is critical to gain a correct and comprehensive understanding of the system, its own state, and the environment. This is of more importance in systems, where contextual information plays a crucial role in the functional operation of the system. In this paper, the authors introduce confidence as a quality metric of observation and leverage it to improve the correct identification of states of a system. To evaluate the impact of this factor on the context-aware monitoring system at hand and to show the generality of the approach, we conduct a series of tests with and without confidence for condition monitoring of an industrial AC motor and an experimental water pipe system. Our experiments show that confidence not only improves the quality of system performance but also simplifies the system architecture and enhances its robustness. These findings support the recent initiatives of paying more attention to observation as an important factor in self-awareness and, consequently, the performance of systems. The proposed system facilitates condition monitoring of various industrial systems and is easily deployable as it does not require a deep domain knowledge.

KEYWORDS:

• Industry 4.0
• monitoring
• model-free
• context-awareness
• self-awareness
• confidence

View correction statement:

Correction

Acknowledgments

We gratefully acknowledge the financial support provided to us by the BMVIT/FFG under the program Production of the Future in the project SAVE (FFG 864883).

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Distance means a difference which can be of any dimension such as performance or success rate difference, geometrical distance, time difference, difference in absolute values, and so forth.

2. We note that the relative distance, $d_{i,j}$, is not to be confused with ${\mathrm{\Delta }}^{\prime }$. While $d_{i,j}$ is the distance of a sample to another sample, ${\mathrm{\Delta }}^{\prime }$ denotes the confidence that a membership assessment is correct.

3. There can be more than four points and three intervals. The fuzzy functions could be much more complex. However, the currently chosen shape has led to satisfactory results in our tests.

4. In fuzzy logic, the complement (negation) of a conjunction is not equal to a disjunction of complemented operands. Hence, in a generic case, $c_b\ne c_n$, although in some specific instances they may be equal.

5. We note that if input or output of the SuO is considered independently from each other, $m$ constitutes the number of input and output variables, respectively.

6. ‘Pt’ stands for platinum and ‘100’ shows the resistance at $0^{\circ }$C in ohms.

7. The smallest number of configurations which worked for all experiments of both case studies was 1987 sets.

8. We note that one scenario constitutes the exception: the anomaly (bearing defect). However, we believe that it would also work unfiltered if we had motor data which would show the motor working well in the beginning and before breaking. Unfortunately, we only have records of the broken motor, and therefore, the vibration signals are changing throughout the recording period. Thus, it is impossible for the SH to recognize a valid state, and the SH only raises an alarm if a valid state was found.

Additional information

Funding

This work was supported by the Bundesministerium für Verkehr, Innovation und Technologie [FFG 864883]; Österreichische Forschungsförderungsgesellschaft [FFG 864883].