ESI’s system-data demonstrator
High-tech systems are nowadays generating a large amount of data every day. The question that we address is how to effectively utilize the emerging big data for the benefit of high-tech industry. To explore the challenges and provide technological solutions, we are developing a system-data demonstrator, which applies different data analysis techniques to various types of operational data. The demonstrator is work-in-progress, and this website shows an overview. The details of its current status can be found here.
An industry showcase
Smart high-tech system-diagnostics with operational data
The demonstrator uses a fictitious machine that has an installed base of high-tech systems all over the world. Currently, it demonstrates how smart system diagnostics can be achieved using available data combined with system engineering knowledge. The demonstrator is built around 5 diagnostic challenges, each of which uses different types of data streams and analysis techniques.
1. Process mining identifies performance bottlenecks in a production line
System-of-systems data from a production line and process mining
A production line usually consists of multiple machines operating in a system-of-systems manner. These machines cooperatively process the same objects, and generate operational data. Process mining is employed to analyse the operational data and derive the process flows of objects on the production line. The performance bottlenecks (e.g., object distributions across machines and/or specific machines) can be identified by analysing the extracted flows (see the screenshot). Given the identified performance bottlenecks, next-step analysis, e.g., for specific machines, are recommended.
2. Unsupervised anomaly detection narrows down problem area
Machine data and anomaly detection
To diagnose a specific machine issue and narrow down the space of causes, we use the operational data generated by the machine components. Basically, anomalies are identified based on the operational data. This is achieved using unsupervised learning techniques, which are able to detect anomalies without efforts from domain experts. The anomaly detection algorithms used in this demonstrator are provided by Yazzoom.
3. Structured knowledge representation guides trouble-shooters through the available data with anomaly detection
Machine data & expert system
To help exploring the huge amount of data available, system engineering knowledge is used. By structuring the system engineering knowledge and automating anomaly detection where possible, users are guided through the data during trouble-shooting. Finally, the likely root-causes are found.
4. Probalistic reasoning with fleet data prioritizes potential root-causes and suggests actions
Fleet data & probabilistic reasoning
The available information of multiple similar machines can be exploited to prioritize the likely root-causes of a single machine. Particularly, the operational data of a fleet of machines provide evidence whether a likely cause identified on one machine also leads to the same issue on other machines. To reason about such fleet data and knowledge, probabilistic reasoning is used to statistically rank the likely causes and recommend further deep-dive diagnosis of the machine (see the screenshot).
5. Metric temporal logic detects and visualizes bottlenecks in highly dependent tasks and shared resources, such as software modules running on contrained hardware
Machine component data & metric temporal logic
The diagnosis of a machine component relies on its operational data, such as the log events. The failures or errors of the component are identified by verifying its performance metrics, such as latency or throughput requirements. This demonstrator provides a glimpse of the TRACE tool developed by ESI. In particular, the metric temporal logic (MTL) is used to formally verify the performance requirements with the log data of the component. If the requirements are not satisfied, violations are visualized, followed by giving an appropriate recommendation for taking actions. Note that the specification of the requirements is described in a dedicated language that engineers can easily understand. The formal language (i.e., the metric temporal logic) backing the analysis is invisible to the engineers.
LIVE DEMO and further information
Please feel free to contact us: Yonghui Li (email@example.com) or Emile van Gerwen (firstname.lastname@example.org).
The details of the demonstrator can be downloaded here.