Integrated Analytics for IIoT Predictive Maintenance using IoT Big Data Cloud Systems
1 like•565 views
The document outlines a framework for integrated analytics in the context of IIoT predictive maintenance, emphasizing the role of IoT cloud systems, big data analytics, and human tasks. It discusses challenges, case studies, and the necessity for human-in-the-loop processes to optimize data collection and analytics. Future work is proposed, focusing on real-world experiments, automatic instrumentation, and enhanced integration with existing enterprise resources.
Integrated Analytics for IIoT Predictive Maintenance using IoT Big Data Cloud Systems
- 1. Integrated Analytics for IIoT Predictive Maintenance using IoT Big Data Cloud Systems Hong-Linh Truong Faculty of Informatics, TU Wien, Austria [email protected] http://rdsea.github.io @linhsolar IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 1
- 2. Outline IoT Cloud systems for predictive maintenance Holistic task analytics generation Integration of big data analytics with human tasks Prototype and examples Conclusions and future work IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 2
- 3. IoT Cloud systems for IIoT predictive maintenance Common parts of IoT Cloud systems for IIoT IoT sensing and connectivity: MQTT, AMQP, FTP, etc. Edge/Fog: lightweighted data analytics/storage Cloud: IoTHub, Apache Nifi, Hadoop, Spark, Kafka, Flink, ElasticSearch, BigQuery, etc. Challenges in the development But software components and their integration are known Key focuses Moving data and aggregating data Combining both streaming analytics and batch analytics for predictive maintenance IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 3
- 4. Case study: IIoT predictive maintenance for Base Transceiver Stations (BTS) Predictive maintenance of BTS IoT, edge and cloud and enterprise/on-premise clouds Complex data types of data Similar to many other cases for maintenance IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 4
- 5. Integrated System and Application Domain Maintenance IIoT and (complex software) system incidents Any incidents in the system can cause problems in data collection and analytics Equipment analytics Maintenance is based on indicators of equipment Analytics results indicate potential incidents of equipment to be maintained. Analytics results of equipment are the output of big/streaming data analysis of equipment status Many types of analytics and context-specific instantiation IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 5
- 6. Integrated needs IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 6 We need both IT and application domain maintenance Integrated System and Application Domain Maintenance But this cannot be done fully automatically by software (even with powerful AI techniques) IIoT with predictive maintenance functions built atop big data analysis and expert capabilities
- 7. Steps IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 7 Algorithms and data analytics detect potential indicators Possible indicators are encapsulated in socalled critical analytics results Critical analytics results are propagated into the right components, which are software or humans, for further consideration
- 8. Tasks suitable for humans Control and validate IoT data collection processes E.g. control the quality of data collection to figure out if it is a problem of the IoT Cloud systems Change and deploy data analytics functions Data analytics functions are strongly dependent on experts Not all analytics need to be deployed and run in advance Avoid cost and maintenance IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 8
- 9. Human-in-the-loop for predictive maintenance Change/configure sensors for data collection Sensor/sensing-as-a-service Control resources for handling/sharing data Elasticity of resources and data sharing Select and deploy suitable data analytics On-demand provisioning of analytics Optimize equipment Remote control and optimization Fix the physical systems of equipment IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 9
- 10. Approach to the support of IIoT predictive maintenance IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 10 IT System incidents (IoT Clouds) Human expertise (IoT Clouds +Domain) Critical analytics results (Domain equipment)
- 11. Holistic task analytics and generation Points of instrumentation for capturing critical analytics results indicating issues of equipment and system incidents impacting the data analytics Human-as-a-service for controlling and supporting data collection and analytics (HumanServiceProvisioning Systems - HSPS) Function-as-a-service (FaaS/serverless) for the integration IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 11
- 12. Integration of big data analytics with human tasks IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 12 Instrumentation of analytics Extensible function catalogs Incident records (for systems and critical analytics results)
- 13. Integration with Human services Human service provisioning systems should be external Tradeoff w.r.t enterprise resources integration (Automatic) scheduling human tasks is complex IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 13 Interactions with third party human service provisioning systems Example with RAHYMS (https://github.com/tuwiendsg/RAHYMS)
- 14. Principles for integration with existing HSPS: Potential human service systems Common systems like: Bots, Slack, OpsGenie, etc. But hard to incorporate application domain knowledge and automatic tasks mapping Important considerations for integrating with other services in IoT Cloud systems, HSPS must provide well-defined APIs, e.g., REST API calls and task structure HSPS must allow domain-specific knowledge to be defined, e.g., rules and human specifications IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 14
- 15. Prototype Data sources: MQTT and Logs from vendor- specific systems for BTS in Vietnam Big data storage: HDFS, BigQuery and Google Storage Platform services: Apache Nifi and RabbitMQT Analytics: Apache Spark & python-based ML Serverless: serverless framework with Google Cloud Functions, Human interaction: RAHYMS. IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 15
- 16. Examples of checking quality of data Using Apache Nifi flows transferring data for maintainance analytics Missing or bad quality of data might trigger issues with equipment Report system incidents based on quality of data IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 16 a PoI in Nifi sending a system incident through RabbitMQ
- 17. Triggering in- depth analytics of several months of historical data based on streaming analytics IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 17
- 18. Functions for problem-to-human task IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 18 pySpark analytics can report problems Task generation Instrument pySpark analytics programs & create a human tasks when there are many alarms determined
- 19. Examples of Professionals and Tasks IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 19 Rules for mapping to expertise RAHYMS: https://github.com/tuwiendsg/RAHYMS
- 20. Examples of Professionals and Tasks IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 20 Messages sent to expert Accept/Reject
- 21. Conclusions and future work IIoT predictive maintenance must consider both systems incidents and application domain incidents in an integrated manner Strong dependences among IoTCloud Systems and analytics of equipment in complex IIoT Human interactions play a key role For both IoT Cloud Systems and equipment analytics In this paper we focus on service engineering aspects for IIoT Integrating IoT Cloud systems, big data and human tasks are pre-requisite for intelligence predictive maintenance IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 21
- 22. Conclusions and future work We introduce an architecture and framework for IIoT predictive maintenance Consider both system incidents and critical analytics results Leverage serverless as a flexible way to integrate big data analytics with human tasks Integrate with external human services. Future work Real world experiments Automatic instrumentation and mapping of analytics Integration of existing enterprise human resources IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 22
- 23. Thanks for your attention! Hong-Linh Truong Faculty of Informatics TU Wien rdsea.github.io IEEE ICII 2018, 22nd Oct 2018, Bellevue, USA 23